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