1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MM_TYPES_H
3 #define _LINUX_MM_TYPES_H
4
5 #include <linux/mm_types_task.h>
6
7 #include <linux/auxvec.h>
8 #include <linux/kref.h>
9 #include <linux/list.h>
10 #include <linux/spinlock.h>
11 #include <linux/rbtree.h>
12 #include <linux/maple_tree.h>
13 #include <linux/rwsem.h>
14 #include <linux/completion.h>
15 #include <linux/cpumask.h>
16 #include <linux/uprobes.h>
17 #include <linux/rcupdate.h>
18 #include <linux/page-flags-layout.h>
19 #include <linux/workqueue.h>
20 #include <linux/seqlock.h>
21 #include <linux/percpu_counter.h>
22
23 #include <asm/mmu.h>
24
25 #ifndef AT_VECTOR_SIZE_ARCH
26 #define AT_VECTOR_SIZE_ARCH 0
27 #endif
28 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
29
30 #define INIT_PASID 0
31
32 struct address_space;
33 struct mem_cgroup;
34
35 /*
36 * Each physical page in the system has a struct page associated with
37 * it to keep track of whatever it is we are using the page for at the
38 * moment. Note that we have no way to track which tasks are using
39 * a page, though if it is a pagecache page, rmap structures can tell us
40 * who is mapping it.
41 *
42 * If you allocate the page using alloc_pages(), you can use some of the
43 * space in struct page for your own purposes. The five words in the main
44 * union are available, except for bit 0 of the first word which must be
45 * kept clear. Many users use this word to store a pointer to an object
46 * which is guaranteed to be aligned. If you use the same storage as
47 * page->mapping, you must restore it to NULL before freeing the page.
48 *
49 * The mapcount field must not be used for own purposes.
50 *
51 * If you want to use the refcount field, it must be used in such a way
52 * that other CPUs temporarily incrementing and then decrementing the
53 * refcount does not cause problems. On receiving the page from
54 * alloc_pages(), the refcount will be positive.
55 *
56 * If you allocate pages of order > 0, you can use some of the fields
57 * in each subpage, but you may need to restore some of their values
58 * afterwards.
59 *
60 * SLUB uses cmpxchg_double() to atomically update its freelist and counters.
61 * That requires that freelist & counters in struct slab be adjacent and
62 * double-word aligned. Because struct slab currently just reinterprets the
63 * bits of struct page, we align all struct pages to double-word boundaries,
64 * and ensure that 'freelist' is aligned within struct slab.
65 */
66 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
67 #define _struct_page_alignment __aligned(2 * sizeof(unsigned long))
68 #else
69 #define _struct_page_alignment __aligned(sizeof(unsigned long))
70 #endif
71
72 struct page {
73 unsigned long flags; /* Atomic flags, some possibly
74 * updated asynchronously */
75 /*
76 * Five words (20/40 bytes) are available in this union.
77 * WARNING: bit 0 of the first word is used for PageTail(). That
78 * means the other users of this union MUST NOT use the bit to
79 * avoid collision and false-positive PageTail().
80 */
81 union {
82 struct { /* Page cache and anonymous pages */
83 /**
84 * @lru: Pageout list, eg. active_list protected by
85 * lruvec->lru_lock. Sometimes used as a generic list
86 * by the page owner.
87 */
88 union {
89 struct list_head lru;
90
91 /* Or, for the Unevictable "LRU list" slot */
92 struct {
93 /* Always even, to negate PageTail */
94 void *__filler;
95 /* Count page's or folio's mlocks */
96 unsigned int mlock_count;
97 };
98
99 /* Or, free page */
100 struct list_head buddy_list;
101 struct list_head pcp_list;
102 };
103 /* See page-flags.h for PAGE_MAPPING_FLAGS */
104 struct address_space *mapping;
105 union {
106 pgoff_t index; /* Our offset within mapping. */
107 unsigned long share; /* share count for fsdax */
108 };
109 /**
110 * @private: Mapping-private opaque data.
111 * Usually used for buffer_heads if PagePrivate.
112 * Used for swp_entry_t if PageSwapCache.
113 * Indicates order in the buddy system if PageBuddy.
114 */
115 unsigned long private;
116 };
117 struct { /* page_pool used by netstack */
118 /**
119 * @pp_magic: magic value to avoid recycling non
120 * page_pool allocated pages.
121 */
122 unsigned long pp_magic;
123 struct page_pool *pp;
124 unsigned long _pp_mapping_pad;
125 unsigned long dma_addr;
126 atomic_long_t pp_ref_count;
127 };
128 struct { /* Tail pages of compound page */
129 unsigned long compound_head; /* Bit zero is set */
130 };
131 struct { /* ZONE_DEVICE pages */
132 /** @pgmap: Points to the hosting device page map. */
133 struct dev_pagemap *pgmap;
134 void *zone_device_data;
135 /*
136 * ZONE_DEVICE private pages are counted as being
137 * mapped so the next 3 words hold the mapping, index,
138 * and private fields from the source anonymous or
139 * page cache page while the page is migrated to device
140 * private memory.
141 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
142 * use the mapping, index, and private fields when
143 * pmem backed DAX files are mapped.
144 */
145 };
146
147 /** @rcu_head: You can use this to free a page by RCU. */
148 struct rcu_head rcu_head;
149 };
150
151 union { /* This union is 4 bytes in size. */
152 /*
153 * For head pages of typed folios, the value stored here
154 * allows for determining what this page is used for. The
155 * tail pages of typed folios will not store a type
156 * (page_type == _mapcount == -1).
157 *
158 * See page-flags.h for a list of page types which are currently
159 * stored here.
160 *
161 * Owners of typed folios may reuse the lower 16 bit of the
162 * head page page_type field after setting the page type,
163 * but must reset these 16 bit to -1 before clearing the
164 * page type.
165 */
166 unsigned int page_type;
167
168 /*
169 * For pages that are part of non-typed folios for which mappings
170 * are tracked via the RMAP, encodes the number of times this page
171 * is directly referenced by a page table.
172 *
173 * Note that the mapcount is always initialized to -1, so that
174 * transitions both from it and to it can be tracked, using
175 * atomic_inc_and_test() and atomic_add_negative(-1).
176 */
177 atomic_t _mapcount;
178 };
179
180 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
181 atomic_t _refcount;
182
183 #ifdef CONFIG_MEMCG
184 unsigned long memcg_data;
185 #elif defined(CONFIG_SLAB_OBJ_EXT)
186 unsigned long _unused_slab_obj_exts;
187 #endif
188
189 /*
190 * On machines where all RAM is mapped into kernel address space,
191 * we can simply calculate the virtual address. On machines with
192 * highmem some memory is mapped into kernel virtual memory
193 * dynamically, so we need a place to store that address.
194 * Note that this field could be 16 bits on x86 ... ;)
195 *
196 * Architectures with slow multiplication can define
197 * WANT_PAGE_VIRTUAL in asm/page.h
198 */
199 #if defined(WANT_PAGE_VIRTUAL)
200 void *virtual; /* Kernel virtual address (NULL if
201 not kmapped, ie. highmem) */
202 #endif /* WANT_PAGE_VIRTUAL */
203
204 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
205 int _last_cpupid;
206 #endif
207
208 #ifdef CONFIG_KMSAN
209 /*
210 * KMSAN metadata for this page:
211 * - shadow page: every bit indicates whether the corresponding
212 * bit of the original page is initialized (0) or not (1);
213 * - origin page: every 4 bytes contain an id of the stack trace
214 * where the uninitialized value was created.
215 */
216 struct page *kmsan_shadow;
217 struct page *kmsan_origin;
218 #endif
219 } _struct_page_alignment;
220
221 /*
222 * struct encoded_page - a nonexistent type marking this pointer
223 *
224 * An 'encoded_page' pointer is a pointer to a regular 'struct page', but
225 * with the low bits of the pointer indicating extra context-dependent
226 * information. Only used in mmu_gather handling, and this acts as a type
227 * system check on that use.
228 *
229 * We only really have two guaranteed bits in general, although you could
230 * play with 'struct page' alignment (see CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
231 * for more.
232 *
233 * Use the supplied helper functions to endcode/decode the pointer and bits.
234 */
235 struct encoded_page;
236
237 #define ENCODED_PAGE_BITS 3ul
238
239 /* Perform rmap removal after we have flushed the TLB. */
240 #define ENCODED_PAGE_BIT_DELAY_RMAP 1ul
241
242 /*
243 * The next item in an encoded_page array is the "nr_pages" argument, specifying
244 * the number of consecutive pages starting from this page, that all belong to
245 * the same folio. For example, "nr_pages" corresponds to the number of folio
246 * references that must be dropped. If this bit is not set, "nr_pages" is
247 * implicitly 1.
248 */
249 #define ENCODED_PAGE_BIT_NR_PAGES_NEXT 2ul
250
encode_page(struct page * page,unsigned long flags)251 static __always_inline struct encoded_page *encode_page(struct page *page, unsigned long flags)
252 {
253 BUILD_BUG_ON(flags > ENCODED_PAGE_BITS);
254 return (struct encoded_page *)(flags | (unsigned long)page);
255 }
256
encoded_page_flags(struct encoded_page * page)257 static inline unsigned long encoded_page_flags(struct encoded_page *page)
258 {
259 return ENCODED_PAGE_BITS & (unsigned long)page;
260 }
261
encoded_page_ptr(struct encoded_page * page)262 static inline struct page *encoded_page_ptr(struct encoded_page *page)
263 {
264 return (struct page *)(~ENCODED_PAGE_BITS & (unsigned long)page);
265 }
266
encode_nr_pages(unsigned long nr)267 static __always_inline struct encoded_page *encode_nr_pages(unsigned long nr)
268 {
269 VM_WARN_ON_ONCE((nr << 2) >> 2 != nr);
270 return (struct encoded_page *)(nr << 2);
271 }
272
encoded_nr_pages(struct encoded_page * page)273 static __always_inline unsigned long encoded_nr_pages(struct encoded_page *page)
274 {
275 return ((unsigned long)page) >> 2;
276 }
277
278 /*
279 * A swap entry has to fit into a "unsigned long", as the entry is hidden
280 * in the "index" field of the swapper address space.
281 */
282 typedef struct {
283 unsigned long val;
284 } swp_entry_t;
285
286 /**
287 * struct folio - Represents a contiguous set of bytes.
288 * @flags: Identical to the page flags.
289 * @lru: Least Recently Used list; tracks how recently this folio was used.
290 * @mlock_count: Number of times this folio has been pinned by mlock().
291 * @mapping: The file this page belongs to, or refers to the anon_vma for
292 * anonymous memory.
293 * @index: Offset within the file, in units of pages. For anonymous memory,
294 * this is the index from the beginning of the mmap.
295 * @private: Filesystem per-folio data (see folio_attach_private()).
296 * @swap: Used for swp_entry_t if folio_test_swapcache().
297 * @_mapcount: Do not access this member directly. Use folio_mapcount() to
298 * find out how many times this folio is mapped by userspace.
299 * @_refcount: Do not access this member directly. Use folio_ref_count()
300 * to find how many references there are to this folio.
301 * @memcg_data: Memory Control Group data.
302 * @virtual: Virtual address in the kernel direct map.
303 * @_last_cpupid: IDs of last CPU and last process that accessed the folio.
304 * @_entire_mapcount: Do not use directly, call folio_entire_mapcount().
305 * @_large_mapcount: Do not use directly, call folio_mapcount().
306 * @_nr_pages_mapped: Do not use outside of rmap and debug code.
307 * @_pincount: Do not use directly, call folio_maybe_dma_pinned().
308 * @_folio_nr_pages: Do not use directly, call folio_nr_pages().
309 * @_hugetlb_subpool: Do not use directly, use accessor in hugetlb.h.
310 * @_hugetlb_cgroup: Do not use directly, use accessor in hugetlb_cgroup.h.
311 * @_hugetlb_cgroup_rsvd: Do not use directly, use accessor in hugetlb_cgroup.h.
312 * @_hugetlb_hwpoison: Do not use directly, call raw_hwp_list_head().
313 * @_deferred_list: Folios to be split under memory pressure.
314 * @_unused_slab_obj_exts: Placeholder to match obj_exts in struct slab.
315 *
316 * A folio is a physically, virtually and logically contiguous set
317 * of bytes. It is a power-of-two in size, and it is aligned to that
318 * same power-of-two. It is at least as large as %PAGE_SIZE. If it is
319 * in the page cache, it is at a file offset which is a multiple of that
320 * power-of-two. It may be mapped into userspace at an address which is
321 * at an arbitrary page offset, but its kernel virtual address is aligned
322 * to its size.
323 */
324 struct folio {
325 /* private: don't document the anon union */
326 union {
327 struct {
328 /* public: */
329 unsigned long flags;
330 union {
331 struct list_head lru;
332 /* private: avoid cluttering the output */
333 struct {
334 void *__filler;
335 /* public: */
336 unsigned int mlock_count;
337 /* private: */
338 };
339 /* public: */
340 };
341 struct address_space *mapping;
342 pgoff_t index;
343 union {
344 void *private;
345 swp_entry_t swap;
346 };
347 atomic_t _mapcount;
348 atomic_t _refcount;
349 #ifdef CONFIG_MEMCG
350 unsigned long memcg_data;
351 #elif defined(CONFIG_SLAB_OBJ_EXT)
352 unsigned long _unused_slab_obj_exts;
353 #endif
354 #if defined(WANT_PAGE_VIRTUAL)
355 void *virtual;
356 #endif
357 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
358 int _last_cpupid;
359 #endif
360 /* private: the union with struct page is transitional */
361 };
362 struct page page;
363 };
364 union {
365 struct {
366 unsigned long _flags_1;
367 unsigned long _head_1;
368 /* public: */
369 atomic_t _large_mapcount;
370 atomic_t _entire_mapcount;
371 atomic_t _nr_pages_mapped;
372 atomic_t _pincount;
373 #ifdef CONFIG_64BIT
374 unsigned int _folio_nr_pages;
375 #endif
376 /* private: the union with struct page is transitional */
377 };
378 struct page __page_1;
379 };
380 union {
381 struct {
382 unsigned long _flags_2;
383 unsigned long _head_2;
384 /* public: */
385 void *_hugetlb_subpool;
386 void *_hugetlb_cgroup;
387 void *_hugetlb_cgroup_rsvd;
388 void *_hugetlb_hwpoison;
389 /* private: the union with struct page is transitional */
390 };
391 struct {
392 unsigned long _flags_2a;
393 unsigned long _head_2a;
394 /* public: */
395 struct list_head _deferred_list;
396 /* private: the union with struct page is transitional */
397 };
398 struct page __page_2;
399 };
400 };
401
402 #define FOLIO_MATCH(pg, fl) \
403 static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
404 FOLIO_MATCH(flags, flags);
405 FOLIO_MATCH(lru, lru);
406 FOLIO_MATCH(mapping, mapping);
407 FOLIO_MATCH(compound_head, lru);
408 FOLIO_MATCH(index, index);
409 FOLIO_MATCH(private, private);
410 FOLIO_MATCH(_mapcount, _mapcount);
411 FOLIO_MATCH(_refcount, _refcount);
412 #ifdef CONFIG_MEMCG
413 FOLIO_MATCH(memcg_data, memcg_data);
414 #endif
415 #if defined(WANT_PAGE_VIRTUAL)
416 FOLIO_MATCH(virtual, virtual);
417 #endif
418 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
419 FOLIO_MATCH(_last_cpupid, _last_cpupid);
420 #endif
421 #undef FOLIO_MATCH
422 #define FOLIO_MATCH(pg, fl) \
423 static_assert(offsetof(struct folio, fl) == \
424 offsetof(struct page, pg) + sizeof(struct page))
425 FOLIO_MATCH(flags, _flags_1);
426 FOLIO_MATCH(compound_head, _head_1);
427 #undef FOLIO_MATCH
428 #define FOLIO_MATCH(pg, fl) \
429 static_assert(offsetof(struct folio, fl) == \
430 offsetof(struct page, pg) + 2 * sizeof(struct page))
431 FOLIO_MATCH(flags, _flags_2);
432 FOLIO_MATCH(compound_head, _head_2);
433 FOLIO_MATCH(flags, _flags_2a);
434 FOLIO_MATCH(compound_head, _head_2a);
435 #undef FOLIO_MATCH
436
437 /**
438 * struct ptdesc - Memory descriptor for page tables.
439 * @__page_flags: Same as page flags. Powerpc only.
440 * @pt_rcu_head: For freeing page table pages.
441 * @pt_list: List of used page tables. Used for s390 and x86.
442 * @_pt_pad_1: Padding that aliases with page's compound head.
443 * @pmd_huge_pte: Protected by ptdesc->ptl, used for THPs.
444 * @__page_mapping: Aliases with page->mapping. Unused for page tables.
445 * @pt_index: Used for s390 gmap.
446 * @pt_mm: Used for x86 pgds.
447 * @pt_frag_refcount: For fragmented page table tracking. Powerpc only.
448 * @_pt_pad_2: Padding to ensure proper alignment.
449 * @ptl: Lock for the page table.
450 * @__page_type: Same as page->page_type. Unused for page tables.
451 * @__page_refcount: Same as page refcount.
452 * @pt_memcg_data: Memcg data. Tracked for page tables here.
453 *
454 * This struct overlays struct page for now. Do not modify without a good
455 * understanding of the issues.
456 */
457 struct ptdesc {
458 unsigned long __page_flags;
459
460 union {
461 struct rcu_head pt_rcu_head;
462 struct list_head pt_list;
463 struct {
464 unsigned long _pt_pad_1;
465 pgtable_t pmd_huge_pte;
466 };
467 };
468 unsigned long __page_mapping;
469
470 union {
471 pgoff_t pt_index;
472 struct mm_struct *pt_mm;
473 atomic_t pt_frag_refcount;
474 };
475
476 union {
477 unsigned long _pt_pad_2;
478 #if ALLOC_SPLIT_PTLOCKS
479 spinlock_t *ptl;
480 #else
481 spinlock_t ptl;
482 #endif
483 };
484 unsigned int __page_type;
485 atomic_t __page_refcount;
486 #ifdef CONFIG_MEMCG
487 unsigned long pt_memcg_data;
488 #endif
489 };
490
491 #define TABLE_MATCH(pg, pt) \
492 static_assert(offsetof(struct page, pg) == offsetof(struct ptdesc, pt))
493 TABLE_MATCH(flags, __page_flags);
494 TABLE_MATCH(compound_head, pt_list);
495 TABLE_MATCH(compound_head, _pt_pad_1);
496 TABLE_MATCH(mapping, __page_mapping);
497 TABLE_MATCH(index, pt_index);
498 TABLE_MATCH(rcu_head, pt_rcu_head);
499 TABLE_MATCH(page_type, __page_type);
500 TABLE_MATCH(_refcount, __page_refcount);
501 #ifdef CONFIG_MEMCG
502 TABLE_MATCH(memcg_data, pt_memcg_data);
503 #endif
504 #undef TABLE_MATCH
505 static_assert(sizeof(struct ptdesc) <= sizeof(struct page));
506
507 #define ptdesc_page(pt) (_Generic((pt), \
508 const struct ptdesc *: (const struct page *)(pt), \
509 struct ptdesc *: (struct page *)(pt)))
510
511 #define ptdesc_folio(pt) (_Generic((pt), \
512 const struct ptdesc *: (const struct folio *)(pt), \
513 struct ptdesc *: (struct folio *)(pt)))
514
515 #define page_ptdesc(p) (_Generic((p), \
516 const struct page *: (const struct ptdesc *)(p), \
517 struct page *: (struct ptdesc *)(p)))
518
519 /*
520 * Used for sizing the vmemmap region on some architectures
521 */
522 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
523
524 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
525 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
526
527 /*
528 * page_private can be used on tail pages. However, PagePrivate is only
529 * checked by the VM on the head page. So page_private on the tail pages
530 * should be used for data that's ancillary to the head page (eg attaching
531 * buffer heads to tail pages after attaching buffer heads to the head page)
532 */
533 #define page_private(page) ((page)->private)
534
set_page_private(struct page * page,unsigned long private)535 static inline void set_page_private(struct page *page, unsigned long private)
536 {
537 page->private = private;
538 }
539
folio_get_private(struct folio * folio)540 static inline void *folio_get_private(struct folio *folio)
541 {
542 return folio->private;
543 }
544
545 struct page_frag_cache {
546 void * va;
547 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
548 __u16 offset;
549 __u16 size;
550 #else
551 __u32 offset;
552 #endif
553 /* we maintain a pagecount bias, so that we dont dirty cache line
554 * containing page->_refcount every time we allocate a fragment.
555 */
556 unsigned int pagecnt_bias;
557 bool pfmemalloc;
558 };
559
560 typedef unsigned long vm_flags_t;
561
562 /*
563 * A region containing a mapping of a non-memory backed file under NOMMU
564 * conditions. These are held in a global tree and are pinned by the VMAs that
565 * map parts of them.
566 */
567 struct vm_region {
568 struct rb_node vm_rb; /* link in global region tree */
569 vm_flags_t vm_flags; /* VMA vm_flags */
570 unsigned long vm_start; /* start address of region */
571 unsigned long vm_end; /* region initialised to here */
572 unsigned long vm_top; /* region allocated to here */
573 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
574 struct file *vm_file; /* the backing file or NULL */
575
576 int vm_usage; /* region usage count (access under nommu_region_sem) */
577 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
578 * this region */
579 };
580
581 #ifdef CONFIG_USERFAULTFD
582 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
583 struct vm_userfaultfd_ctx {
584 struct userfaultfd_ctx *ctx;
585 };
586 #else /* CONFIG_USERFAULTFD */
587 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
588 struct vm_userfaultfd_ctx {};
589 #endif /* CONFIG_USERFAULTFD */
590
591 struct anon_vma_name {
592 struct kref kref;
593 /* The name needs to be at the end because it is dynamically sized. */
594 char name[];
595 };
596
597 #ifdef CONFIG_ANON_VMA_NAME
598 /*
599 * mmap_lock should be read-locked when calling anon_vma_name(). Caller should
600 * either keep holding the lock while using the returned pointer or it should
601 * raise anon_vma_name refcount before releasing the lock.
602 */
603 struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma);
604 struct anon_vma_name *anon_vma_name_alloc(const char *name);
605 void anon_vma_name_free(struct kref *kref);
606 #else /* CONFIG_ANON_VMA_NAME */
anon_vma_name(struct vm_area_struct * vma)607 static inline struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma)
608 {
609 return NULL;
610 }
611
anon_vma_name_alloc(const char * name)612 static inline struct anon_vma_name *anon_vma_name_alloc(const char *name)
613 {
614 return NULL;
615 }
616 #endif
617
618 struct vma_lock {
619 struct rw_semaphore lock;
620 };
621
622 struct vma_numab_state {
623 /*
624 * Initialised as time in 'jiffies' after which VMA
625 * should be scanned. Delays first scan of new VMA by at
626 * least sysctl_numa_balancing_scan_delay:
627 */
628 unsigned long next_scan;
629
630 /*
631 * Time in jiffies when pids_active[] is reset to
632 * detect phase change behaviour:
633 */
634 unsigned long pids_active_reset;
635
636 /*
637 * Approximate tracking of PIDs that trapped a NUMA hinting
638 * fault. May produce false positives due to hash collisions.
639 *
640 * [0] Previous PID tracking
641 * [1] Current PID tracking
642 *
643 * Window moves after next_pid_reset has expired approximately
644 * every VMA_PID_RESET_PERIOD jiffies:
645 */
646 unsigned long pids_active[2];
647
648 /* MM scan sequence ID when scan first started after VMA creation */
649 int start_scan_seq;
650
651 /*
652 * MM scan sequence ID when the VMA was last completely scanned.
653 * A VMA is not eligible for scanning if prev_scan_seq == numa_scan_seq
654 */
655 int prev_scan_seq;
656 };
657
658 /*
659 * This struct describes a virtual memory area. There is one of these
660 * per VM-area/task. A VM area is any part of the process virtual memory
661 * space that has a special rule for the page-fault handlers (ie a shared
662 * library, the executable area etc).
663 */
664 struct vm_area_struct {
665 /* The first cache line has the info for VMA tree walking. */
666
667 union {
668 struct {
669 /* VMA covers [vm_start; vm_end) addresses within mm */
670 unsigned long vm_start;
671 unsigned long vm_end;
672 };
673 #ifdef CONFIG_PER_VMA_LOCK
674 struct rcu_head vm_rcu; /* Used for deferred freeing. */
675 #endif
676 };
677
678 struct mm_struct *vm_mm; /* The address space we belong to. */
679 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
680
681 /*
682 * Flags, see mm.h.
683 * To modify use vm_flags_{init|reset|set|clear|mod} functions.
684 */
685 union {
686 const vm_flags_t vm_flags;
687 vm_flags_t __private __vm_flags;
688 };
689
690 #ifdef CONFIG_PER_VMA_LOCK
691 /* Flag to indicate areas detached from the mm->mm_mt tree */
692 bool detached;
693
694 /*
695 * Can only be written (using WRITE_ONCE()) while holding both:
696 * - mmap_lock (in write mode)
697 * - vm_lock->lock (in write mode)
698 * Can be read reliably while holding one of:
699 * - mmap_lock (in read or write mode)
700 * - vm_lock->lock (in read or write mode)
701 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
702 * while holding nothing (except RCU to keep the VMA struct allocated).
703 *
704 * This sequence counter is explicitly allowed to overflow; sequence
705 * counter reuse can only lead to occasional unnecessary use of the
706 * slowpath.
707 */
708 int vm_lock_seq;
709 struct vma_lock *vm_lock;
710 #endif
711
712 /*
713 * For areas with an address space and backing store,
714 * linkage into the address_space->i_mmap interval tree.
715 *
716 */
717 struct {
718 struct rb_node rb;
719 unsigned long rb_subtree_last;
720 } shared;
721
722 /*
723 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
724 * list, after a COW of one of the file pages. A MAP_SHARED vma
725 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
726 * or brk vma (with NULL file) can only be in an anon_vma list.
727 */
728 struct list_head anon_vma_chain; /* Serialized by mmap_lock &
729 * page_table_lock */
730 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
731
732 /* Function pointers to deal with this struct. */
733 const struct vm_operations_struct *vm_ops;
734
735 /* Information about our backing store: */
736 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
737 units */
738 struct file * vm_file; /* File we map to (can be NULL). */
739 void * vm_private_data; /* was vm_pte (shared mem) */
740
741 #ifdef CONFIG_ANON_VMA_NAME
742 /*
743 * For private and shared anonymous mappings, a pointer to a null
744 * terminated string containing the name given to the vma, or NULL if
745 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access.
746 */
747 struct anon_vma_name *anon_name;
748 #endif
749 #ifdef CONFIG_SWAP
750 atomic_long_t swap_readahead_info;
751 #endif
752 #ifndef CONFIG_MMU
753 struct vm_region *vm_region; /* NOMMU mapping region */
754 #endif
755 #ifdef CONFIG_NUMA
756 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
757 #endif
758 #ifdef CONFIG_NUMA_BALANCING
759 struct vma_numab_state *numab_state; /* NUMA Balancing state */
760 #endif
761 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
762 } __randomize_layout;
763
764 #ifdef CONFIG_NUMA
765 #define vma_policy(vma) ((vma)->vm_policy)
766 #else
767 #define vma_policy(vma) NULL
768 #endif
769
770 #ifdef CONFIG_SCHED_MM_CID
771 struct mm_cid {
772 u64 time;
773 int cid;
774 };
775 #endif
776
777 struct kioctx_table;
778 struct iommu_mm_data;
779 struct mm_struct {
780 struct {
781 /*
782 * Fields which are often written to are placed in a separate
783 * cache line.
784 */
785 struct {
786 /**
787 * @mm_count: The number of references to &struct
788 * mm_struct (@mm_users count as 1).
789 *
790 * Use mmgrab()/mmdrop() to modify. When this drops to
791 * 0, the &struct mm_struct is freed.
792 */
793 atomic_t mm_count;
794 } ____cacheline_aligned_in_smp;
795
796 struct maple_tree mm_mt;
797
798 unsigned long mmap_base; /* base of mmap area */
799 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
800 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
801 /* Base addresses for compatible mmap() */
802 unsigned long mmap_compat_base;
803 unsigned long mmap_compat_legacy_base;
804 #endif
805 unsigned long task_size; /* size of task vm space */
806 pgd_t * pgd;
807
808 #ifdef CONFIG_MEMBARRIER
809 /**
810 * @membarrier_state: Flags controlling membarrier behavior.
811 *
812 * This field is close to @pgd to hopefully fit in the same
813 * cache-line, which needs to be touched by switch_mm().
814 */
815 atomic_t membarrier_state;
816 #endif
817
818 /**
819 * @mm_users: The number of users including userspace.
820 *
821 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
822 * drops to 0 (i.e. when the task exits and there are no other
823 * temporary reference holders), we also release a reference on
824 * @mm_count (which may then free the &struct mm_struct if
825 * @mm_count also drops to 0).
826 */
827 atomic_t mm_users;
828
829 #ifdef CONFIG_SCHED_MM_CID
830 /**
831 * @pcpu_cid: Per-cpu current cid.
832 *
833 * Keep track of the currently allocated mm_cid for each cpu.
834 * The per-cpu mm_cid values are serialized by their respective
835 * runqueue locks.
836 */
837 struct mm_cid __percpu *pcpu_cid;
838 /*
839 * @mm_cid_next_scan: Next mm_cid scan (in jiffies).
840 *
841 * When the next mm_cid scan is due (in jiffies).
842 */
843 unsigned long mm_cid_next_scan;
844 #endif
845 #ifdef CONFIG_MMU
846 atomic_long_t pgtables_bytes; /* size of all page tables */
847 #endif
848 int map_count; /* number of VMAs */
849
850 spinlock_t page_table_lock; /* Protects page tables and some
851 * counters
852 */
853 /*
854 * With some kernel config, the current mmap_lock's offset
855 * inside 'mm_struct' is at 0x120, which is very optimal, as
856 * its two hot fields 'count' and 'owner' sit in 2 different
857 * cachelines, and when mmap_lock is highly contended, both
858 * of the 2 fields will be accessed frequently, current layout
859 * will help to reduce cache bouncing.
860 *
861 * So please be careful with adding new fields before
862 * mmap_lock, which can easily push the 2 fields into one
863 * cacheline.
864 */
865 struct rw_semaphore mmap_lock;
866
867 struct list_head mmlist; /* List of maybe swapped mm's. These
868 * are globally strung together off
869 * init_mm.mmlist, and are protected
870 * by mmlist_lock
871 */
872 #ifdef CONFIG_PER_VMA_LOCK
873 /*
874 * This field has lock-like semantics, meaning it is sometimes
875 * accessed with ACQUIRE/RELEASE semantics.
876 * Roughly speaking, incrementing the sequence number is
877 * equivalent to releasing locks on VMAs; reading the sequence
878 * number can be part of taking a read lock on a VMA.
879 *
880 * Can be modified under write mmap_lock using RELEASE
881 * semantics.
882 * Can be read with no other protection when holding write
883 * mmap_lock.
884 * Can be read with ACQUIRE semantics if not holding write
885 * mmap_lock.
886 */
887 int mm_lock_seq;
888 #endif
889
890
891 unsigned long hiwater_rss; /* High-watermark of RSS usage */
892 unsigned long hiwater_vm; /* High-water virtual memory usage */
893
894 unsigned long total_vm; /* Total pages mapped */
895 unsigned long locked_vm; /* Pages that have PG_mlocked set */
896 atomic64_t pinned_vm; /* Refcount permanently increased */
897 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
898 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
899 unsigned long stack_vm; /* VM_STACK */
900 unsigned long def_flags;
901
902 /**
903 * @write_protect_seq: Locked when any thread is write
904 * protecting pages mapped by this mm to enforce a later COW,
905 * for instance during page table copying for fork().
906 */
907 seqcount_t write_protect_seq;
908
909 spinlock_t arg_lock; /* protect the below fields */
910
911 unsigned long start_code, end_code, start_data, end_data;
912 unsigned long start_brk, brk, start_stack;
913 unsigned long arg_start, arg_end, env_start, env_end;
914
915 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
916
917 struct percpu_counter rss_stat[NR_MM_COUNTERS];
918
919 struct linux_binfmt *binfmt;
920
921 /* Architecture-specific MM context */
922 mm_context_t context;
923
924 unsigned long flags; /* Must use atomic bitops to access */
925
926 #ifdef CONFIG_AIO
927 spinlock_t ioctx_lock;
928 struct kioctx_table __rcu *ioctx_table;
929 #endif
930 #ifdef CONFIG_MEMCG
931 /*
932 * "owner" points to a task that is regarded as the canonical
933 * user/owner of this mm. All of the following must be true in
934 * order for it to be changed:
935 *
936 * current == mm->owner
937 * current->mm != mm
938 * new_owner->mm == mm
939 * new_owner->alloc_lock is held
940 */
941 struct task_struct __rcu *owner;
942 #endif
943 struct user_namespace *user_ns;
944
945 /* store ref to file /proc/<pid>/exe symlink points to */
946 struct file __rcu *exe_file;
947 #ifdef CONFIG_MMU_NOTIFIER
948 struct mmu_notifier_subscriptions *notifier_subscriptions;
949 #endif
950 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
951 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
952 #endif
953 #ifdef CONFIG_NUMA_BALANCING
954 /*
955 * numa_next_scan is the next time that PTEs will be remapped
956 * PROT_NONE to trigger NUMA hinting faults; such faults gather
957 * statistics and migrate pages to new nodes if necessary.
958 */
959 unsigned long numa_next_scan;
960
961 /* Restart point for scanning and remapping PTEs. */
962 unsigned long numa_scan_offset;
963
964 /* numa_scan_seq prevents two threads remapping PTEs. */
965 int numa_scan_seq;
966 #endif
967 /*
968 * An operation with batched TLB flushing is going on. Anything
969 * that can move process memory needs to flush the TLB when
970 * moving a PROT_NONE mapped page.
971 */
972 atomic_t tlb_flush_pending;
973 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
974 /* See flush_tlb_batched_pending() */
975 atomic_t tlb_flush_batched;
976 #endif
977 struct uprobes_state uprobes_state;
978 #ifdef CONFIG_PREEMPT_RT
979 struct rcu_head delayed_drop;
980 #endif
981 #ifdef CONFIG_HUGETLB_PAGE
982 atomic_long_t hugetlb_usage;
983 #endif
984 struct work_struct async_put_work;
985
986 #ifdef CONFIG_IOMMU_MM_DATA
987 struct iommu_mm_data *iommu_mm;
988 #endif
989 #ifdef CONFIG_KSM
990 /*
991 * Represent how many pages of this process are involved in KSM
992 * merging (not including ksm_zero_pages).
993 */
994 unsigned long ksm_merging_pages;
995 /*
996 * Represent how many pages are checked for ksm merging
997 * including merged and not merged.
998 */
999 unsigned long ksm_rmap_items;
1000 /*
1001 * Represent how many empty pages are merged with kernel zero
1002 * pages when enabling KSM use_zero_pages.
1003 */
1004 atomic_long_t ksm_zero_pages;
1005 #endif /* CONFIG_KSM */
1006 #ifdef CONFIG_LRU_GEN_WALKS_MMU
1007 struct {
1008 /* this mm_struct is on lru_gen_mm_list */
1009 struct list_head list;
1010 /*
1011 * Set when switching to this mm_struct, as a hint of
1012 * whether it has been used since the last time per-node
1013 * page table walkers cleared the corresponding bits.
1014 */
1015 unsigned long bitmap;
1016 #ifdef CONFIG_MEMCG
1017 /* points to the memcg of "owner" above */
1018 struct mem_cgroup *memcg;
1019 #endif
1020 } lru_gen;
1021 #endif /* CONFIG_LRU_GEN_WALKS_MMU */
1022 } __randomize_layout;
1023
1024 /*
1025 * The mm_cpumask needs to be at the end of mm_struct, because it
1026 * is dynamically sized based on nr_cpu_ids.
1027 */
1028 unsigned long cpu_bitmap[];
1029 };
1030
1031 #define MM_MT_FLAGS (MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN | \
1032 MT_FLAGS_USE_RCU)
1033 extern struct mm_struct init_mm;
1034
1035 /* Pointer magic because the dynamic array size confuses some compilers. */
mm_init_cpumask(struct mm_struct * mm)1036 static inline void mm_init_cpumask(struct mm_struct *mm)
1037 {
1038 unsigned long cpu_bitmap = (unsigned long)mm;
1039
1040 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
1041 cpumask_clear((struct cpumask *)cpu_bitmap);
1042 }
1043
1044 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
mm_cpumask(struct mm_struct * mm)1045 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
1046 {
1047 return (struct cpumask *)&mm->cpu_bitmap;
1048 }
1049
1050 #ifdef CONFIG_LRU_GEN
1051
1052 struct lru_gen_mm_list {
1053 /* mm_struct list for page table walkers */
1054 struct list_head fifo;
1055 /* protects the list above */
1056 spinlock_t lock;
1057 };
1058
1059 #endif /* CONFIG_LRU_GEN */
1060
1061 #ifdef CONFIG_LRU_GEN_WALKS_MMU
1062
1063 void lru_gen_add_mm(struct mm_struct *mm);
1064 void lru_gen_del_mm(struct mm_struct *mm);
1065 void lru_gen_migrate_mm(struct mm_struct *mm);
1066
lru_gen_init_mm(struct mm_struct * mm)1067 static inline void lru_gen_init_mm(struct mm_struct *mm)
1068 {
1069 INIT_LIST_HEAD(&mm->lru_gen.list);
1070 mm->lru_gen.bitmap = 0;
1071 #ifdef CONFIG_MEMCG
1072 mm->lru_gen.memcg = NULL;
1073 #endif
1074 }
1075
lru_gen_use_mm(struct mm_struct * mm)1076 static inline void lru_gen_use_mm(struct mm_struct *mm)
1077 {
1078 /*
1079 * When the bitmap is set, page reclaim knows this mm_struct has been
1080 * used since the last time it cleared the bitmap. So it might be worth
1081 * walking the page tables of this mm_struct to clear the accessed bit.
1082 */
1083 WRITE_ONCE(mm->lru_gen.bitmap, -1);
1084 }
1085
1086 #else /* !CONFIG_LRU_GEN_WALKS_MMU */
1087
lru_gen_add_mm(struct mm_struct * mm)1088 static inline void lru_gen_add_mm(struct mm_struct *mm)
1089 {
1090 }
1091
lru_gen_del_mm(struct mm_struct * mm)1092 static inline void lru_gen_del_mm(struct mm_struct *mm)
1093 {
1094 }
1095
lru_gen_migrate_mm(struct mm_struct * mm)1096 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
1097 {
1098 }
1099
lru_gen_init_mm(struct mm_struct * mm)1100 static inline void lru_gen_init_mm(struct mm_struct *mm)
1101 {
1102 }
1103
lru_gen_use_mm(struct mm_struct * mm)1104 static inline void lru_gen_use_mm(struct mm_struct *mm)
1105 {
1106 }
1107
1108 #endif /* CONFIG_LRU_GEN_WALKS_MMU */
1109
1110 struct vma_iterator {
1111 struct ma_state mas;
1112 };
1113
1114 #define VMA_ITERATOR(name, __mm, __addr) \
1115 struct vma_iterator name = { \
1116 .mas = { \
1117 .tree = &(__mm)->mm_mt, \
1118 .index = __addr, \
1119 .node = NULL, \
1120 .status = ma_start, \
1121 }, \
1122 }
1123
vma_iter_init(struct vma_iterator * vmi,struct mm_struct * mm,unsigned long addr)1124 static inline void vma_iter_init(struct vma_iterator *vmi,
1125 struct mm_struct *mm, unsigned long addr)
1126 {
1127 mas_init(&vmi->mas, &mm->mm_mt, addr);
1128 }
1129
1130 #ifdef CONFIG_SCHED_MM_CID
1131
1132 enum mm_cid_state {
1133 MM_CID_UNSET = -1U, /* Unset state has lazy_put flag set. */
1134 MM_CID_LAZY_PUT = (1U << 31),
1135 };
1136
mm_cid_is_unset(int cid)1137 static inline bool mm_cid_is_unset(int cid)
1138 {
1139 return cid == MM_CID_UNSET;
1140 }
1141
mm_cid_is_lazy_put(int cid)1142 static inline bool mm_cid_is_lazy_put(int cid)
1143 {
1144 return !mm_cid_is_unset(cid) && (cid & MM_CID_LAZY_PUT);
1145 }
1146
mm_cid_is_valid(int cid)1147 static inline bool mm_cid_is_valid(int cid)
1148 {
1149 return !(cid & MM_CID_LAZY_PUT);
1150 }
1151
mm_cid_set_lazy_put(int cid)1152 static inline int mm_cid_set_lazy_put(int cid)
1153 {
1154 return cid | MM_CID_LAZY_PUT;
1155 }
1156
mm_cid_clear_lazy_put(int cid)1157 static inline int mm_cid_clear_lazy_put(int cid)
1158 {
1159 return cid & ~MM_CID_LAZY_PUT;
1160 }
1161
1162 /* Accessor for struct mm_struct's cidmask. */
mm_cidmask(struct mm_struct * mm)1163 static inline cpumask_t *mm_cidmask(struct mm_struct *mm)
1164 {
1165 unsigned long cid_bitmap = (unsigned long)mm;
1166
1167 cid_bitmap += offsetof(struct mm_struct, cpu_bitmap);
1168 /* Skip cpu_bitmap */
1169 cid_bitmap += cpumask_size();
1170 return (struct cpumask *)cid_bitmap;
1171 }
1172
mm_init_cid(struct mm_struct * mm)1173 static inline void mm_init_cid(struct mm_struct *mm)
1174 {
1175 int i;
1176
1177 for_each_possible_cpu(i) {
1178 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, i);
1179
1180 pcpu_cid->cid = MM_CID_UNSET;
1181 pcpu_cid->time = 0;
1182 }
1183 cpumask_clear(mm_cidmask(mm));
1184 }
1185
mm_alloc_cid_noprof(struct mm_struct * mm)1186 static inline int mm_alloc_cid_noprof(struct mm_struct *mm)
1187 {
1188 mm->pcpu_cid = alloc_percpu_noprof(struct mm_cid);
1189 if (!mm->pcpu_cid)
1190 return -ENOMEM;
1191 mm_init_cid(mm);
1192 return 0;
1193 }
1194 #define mm_alloc_cid(...) alloc_hooks(mm_alloc_cid_noprof(__VA_ARGS__))
1195
mm_destroy_cid(struct mm_struct * mm)1196 static inline void mm_destroy_cid(struct mm_struct *mm)
1197 {
1198 free_percpu(mm->pcpu_cid);
1199 mm->pcpu_cid = NULL;
1200 }
1201
mm_cid_size(void)1202 static inline unsigned int mm_cid_size(void)
1203 {
1204 return cpumask_size();
1205 }
1206 #else /* CONFIG_SCHED_MM_CID */
mm_init_cid(struct mm_struct * mm)1207 static inline void mm_init_cid(struct mm_struct *mm) { }
mm_alloc_cid(struct mm_struct * mm)1208 static inline int mm_alloc_cid(struct mm_struct *mm) { return 0; }
mm_destroy_cid(struct mm_struct * mm)1209 static inline void mm_destroy_cid(struct mm_struct *mm) { }
mm_cid_size(void)1210 static inline unsigned int mm_cid_size(void)
1211 {
1212 return 0;
1213 }
1214 #endif /* CONFIG_SCHED_MM_CID */
1215
1216 struct mmu_gather;
1217 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
1218 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
1219 extern void tlb_finish_mmu(struct mmu_gather *tlb);
1220
1221 struct vm_fault;
1222
1223 /**
1224 * typedef vm_fault_t - Return type for page fault handlers.
1225 *
1226 * Page fault handlers return a bitmask of %VM_FAULT values.
1227 */
1228 typedef __bitwise unsigned int vm_fault_t;
1229
1230 /**
1231 * enum vm_fault_reason - Page fault handlers return a bitmask of
1232 * these values to tell the core VM what happened when handling the
1233 * fault. Used to decide whether a process gets delivered SIGBUS or
1234 * just gets major/minor fault counters bumped up.
1235 *
1236 * @VM_FAULT_OOM: Out Of Memory
1237 * @VM_FAULT_SIGBUS: Bad access
1238 * @VM_FAULT_MAJOR: Page read from storage
1239 * @VM_FAULT_HWPOISON: Hit poisoned small page
1240 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
1241 * in upper bits
1242 * @VM_FAULT_SIGSEGV: segmentation fault
1243 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
1244 * @VM_FAULT_LOCKED: ->fault locked the returned page
1245 * @VM_FAULT_RETRY: ->fault blocked, must retry
1246 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
1247 * @VM_FAULT_DONE_COW: ->fault has fully handled COW
1248 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
1249 * fsync() to complete (for synchronous page faults
1250 * in DAX)
1251 * @VM_FAULT_COMPLETED: ->fault completed, meanwhile mmap lock released
1252 * @VM_FAULT_HINDEX_MASK: mask HINDEX value
1253 *
1254 */
1255 enum vm_fault_reason {
1256 VM_FAULT_OOM = (__force vm_fault_t)0x000001,
1257 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
1258 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
1259 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
1260 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
1261 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
1262 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
1263 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
1264 VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
1265 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
1266 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
1267 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
1268 VM_FAULT_COMPLETED = (__force vm_fault_t)0x004000,
1269 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
1270 };
1271
1272 /* Encode hstate index for a hwpoisoned large page */
1273 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
1274 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
1275
1276 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
1277 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
1278 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
1279
1280 #define VM_FAULT_RESULT_TRACE \
1281 { VM_FAULT_OOM, "OOM" }, \
1282 { VM_FAULT_SIGBUS, "SIGBUS" }, \
1283 { VM_FAULT_MAJOR, "MAJOR" }, \
1284 { VM_FAULT_HWPOISON, "HWPOISON" }, \
1285 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
1286 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
1287 { VM_FAULT_NOPAGE, "NOPAGE" }, \
1288 { VM_FAULT_LOCKED, "LOCKED" }, \
1289 { VM_FAULT_RETRY, "RETRY" }, \
1290 { VM_FAULT_FALLBACK, "FALLBACK" }, \
1291 { VM_FAULT_DONE_COW, "DONE_COW" }, \
1292 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }, \
1293 { VM_FAULT_COMPLETED, "COMPLETED" }
1294
1295 struct vm_special_mapping {
1296 const char *name; /* The name, e.g. "[vdso]". */
1297
1298 /*
1299 * If .fault is not provided, this points to a
1300 * NULL-terminated array of pages that back the special mapping.
1301 *
1302 * This must not be NULL unless .fault is provided.
1303 */
1304 struct page **pages;
1305
1306 /*
1307 * If non-NULL, then this is called to resolve page faults
1308 * on the special mapping. If used, .pages is not checked.
1309 */
1310 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
1311 struct vm_area_struct *vma,
1312 struct vm_fault *vmf);
1313
1314 int (*mremap)(const struct vm_special_mapping *sm,
1315 struct vm_area_struct *new_vma);
1316 };
1317
1318 enum tlb_flush_reason {
1319 TLB_FLUSH_ON_TASK_SWITCH,
1320 TLB_REMOTE_SHOOTDOWN,
1321 TLB_LOCAL_SHOOTDOWN,
1322 TLB_LOCAL_MM_SHOOTDOWN,
1323 TLB_REMOTE_SEND_IPI,
1324 NR_TLB_FLUSH_REASONS,
1325 };
1326
1327 /**
1328 * enum fault_flag - Fault flag definitions.
1329 * @FAULT_FLAG_WRITE: Fault was a write fault.
1330 * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
1331 * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
1332 * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
1333 * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
1334 * @FAULT_FLAG_TRIED: The fault has been tried once.
1335 * @FAULT_FLAG_USER: The fault originated in userspace.
1336 * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
1337 * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
1338 * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
1339 * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a
1340 * COW mapping, making sure that an exclusive anon page is
1341 * mapped after the fault.
1342 * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached.
1343 * We should only access orig_pte if this flag set.
1344 * @FAULT_FLAG_VMA_LOCK: The fault is handled under VMA lock.
1345 *
1346 * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
1347 * whether we would allow page faults to retry by specifying these two
1348 * fault flags correctly. Currently there can be three legal combinations:
1349 *
1350 * (a) ALLOW_RETRY and !TRIED: this means the page fault allows retry, and
1351 * this is the first try
1352 *
1353 * (b) ALLOW_RETRY and TRIED: this means the page fault allows retry, and
1354 * we've already tried at least once
1355 *
1356 * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
1357 *
1358 * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
1359 * be used. Note that page faults can be allowed to retry for multiple times,
1360 * in which case we'll have an initial fault with flags (a) then later on
1361 * continuous faults with flags (b). We should always try to detect pending
1362 * signals before a retry to make sure the continuous page faults can still be
1363 * interrupted if necessary.
1364 *
1365 * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal.
1366 * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when
1367 * applied to mappings that are not COW mappings.
1368 */
1369 enum fault_flag {
1370 FAULT_FLAG_WRITE = 1 << 0,
1371 FAULT_FLAG_MKWRITE = 1 << 1,
1372 FAULT_FLAG_ALLOW_RETRY = 1 << 2,
1373 FAULT_FLAG_RETRY_NOWAIT = 1 << 3,
1374 FAULT_FLAG_KILLABLE = 1 << 4,
1375 FAULT_FLAG_TRIED = 1 << 5,
1376 FAULT_FLAG_USER = 1 << 6,
1377 FAULT_FLAG_REMOTE = 1 << 7,
1378 FAULT_FLAG_INSTRUCTION = 1 << 8,
1379 FAULT_FLAG_INTERRUPTIBLE = 1 << 9,
1380 FAULT_FLAG_UNSHARE = 1 << 10,
1381 FAULT_FLAG_ORIG_PTE_VALID = 1 << 11,
1382 FAULT_FLAG_VMA_LOCK = 1 << 12,
1383 };
1384
1385 typedef unsigned int __bitwise zap_flags_t;
1386
1387 /* Flags for clear_young_dirty_ptes(). */
1388 typedef int __bitwise cydp_t;
1389
1390 /* Clear the access bit */
1391 #define CYDP_CLEAR_YOUNG ((__force cydp_t)BIT(0))
1392
1393 /* Clear the dirty bit */
1394 #define CYDP_CLEAR_DIRTY ((__force cydp_t)BIT(1))
1395
1396 /*
1397 * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each
1398 * other. Here is what they mean, and how to use them:
1399 *
1400 *
1401 * FIXME: For pages which are part of a filesystem, mappings are subject to the
1402 * lifetime enforced by the filesystem and we need guarantees that longterm
1403 * users like RDMA and V4L2 only establish mappings which coordinate usage with
1404 * the filesystem. Ideas for this coordination include revoking the longterm
1405 * pin, delaying writeback, bounce buffer page writeback, etc. As FS DAX was
1406 * added after the problem with filesystems was found FS DAX VMAs are
1407 * specifically failed. Filesystem pages are still subject to bugs and use of
1408 * FOLL_LONGTERM should be avoided on those pages.
1409 *
1410 * In the CMA case: long term pins in a CMA region would unnecessarily fragment
1411 * that region. And so, CMA attempts to migrate the page before pinning, when
1412 * FOLL_LONGTERM is specified.
1413 *
1414 * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount,
1415 * but an additional pin counting system) will be invoked. This is intended for
1416 * anything that gets a page reference and then touches page data (for example,
1417 * Direct IO). This lets the filesystem know that some non-file-system entity is
1418 * potentially changing the pages' data. In contrast to FOLL_GET (whose pages
1419 * are released via put_page()), FOLL_PIN pages must be released, ultimately, by
1420 * a call to unpin_user_page().
1421 *
1422 * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different
1423 * and separate refcounting mechanisms, however, and that means that each has
1424 * its own acquire and release mechanisms:
1425 *
1426 * FOLL_GET: get_user_pages*() to acquire, and put_page() to release.
1427 *
1428 * FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release.
1429 *
1430 * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call.
1431 * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based
1432 * calls applied to them, and that's perfectly OK. This is a constraint on the
1433 * callers, not on the pages.)
1434 *
1435 * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never
1436 * directly by the caller. That's in order to help avoid mismatches when
1437 * releasing pages: get_user_pages*() pages must be released via put_page(),
1438 * while pin_user_pages*() pages must be released via unpin_user_page().
1439 *
1440 * Please see Documentation/core-api/pin_user_pages.rst for more information.
1441 */
1442
1443 enum {
1444 /* check pte is writable */
1445 FOLL_WRITE = 1 << 0,
1446 /* do get_page on page */
1447 FOLL_GET = 1 << 1,
1448 /* give error on hole if it would be zero */
1449 FOLL_DUMP = 1 << 2,
1450 /* get_user_pages read/write w/o permission */
1451 FOLL_FORCE = 1 << 3,
1452 /*
1453 * if a disk transfer is needed, start the IO and return without waiting
1454 * upon it
1455 */
1456 FOLL_NOWAIT = 1 << 4,
1457 /* do not fault in pages */
1458 FOLL_NOFAULT = 1 << 5,
1459 /* check page is hwpoisoned */
1460 FOLL_HWPOISON = 1 << 6,
1461 /* don't do file mappings */
1462 FOLL_ANON = 1 << 7,
1463 /*
1464 * FOLL_LONGTERM indicates that the page will be held for an indefinite
1465 * time period _often_ under userspace control. This is in contrast to
1466 * iov_iter_get_pages(), whose usages are transient.
1467 */
1468 FOLL_LONGTERM = 1 << 8,
1469 /* split huge pmd before returning */
1470 FOLL_SPLIT_PMD = 1 << 9,
1471 /* allow returning PCI P2PDMA pages */
1472 FOLL_PCI_P2PDMA = 1 << 10,
1473 /* allow interrupts from generic signals */
1474 FOLL_INTERRUPTIBLE = 1 << 11,
1475 /*
1476 * Always honor (trigger) NUMA hinting faults.
1477 *
1478 * FOLL_WRITE implicitly honors NUMA hinting faults because a
1479 * PROT_NONE-mapped page is not writable (exceptions with FOLL_FORCE
1480 * apply). get_user_pages_fast_only() always implicitly honors NUMA
1481 * hinting faults.
1482 */
1483 FOLL_HONOR_NUMA_FAULT = 1 << 12,
1484
1485 /* See also internal only FOLL flags in mm/internal.h */
1486 };
1487
1488 #endif /* _LINUX_MM_TYPES_H */
1489