1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MM_TYPES_H
3 #define _LINUX_MM_TYPES_H
4
5 #include <linux/mm_types_task.h>
6
7 #include <linux/auxvec.h>
8 #include <linux/kref.h>
9 #include <linux/list.h>
10 #include <linux/spinlock.h>
11 #include <linux/rbtree.h>
12 #include <linux/maple_tree.h>
13 #include <linux/rwsem.h>
14 #include <linux/completion.h>
15 #include <linux/cpumask.h>
16 #include <linux/uprobes.h>
17 #include <linux/rcupdate.h>
18 #include <linux/page-flags-layout.h>
19 #include <linux/workqueue.h>
20 #include <linux/seqlock.h>
21 #include <linux/percpu_counter.h>
22
23 #include <asm/mmu.h>
24
25 #ifndef AT_VECTOR_SIZE_ARCH
26 #define AT_VECTOR_SIZE_ARCH 0
27 #endif
28 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
29
30 #define INIT_PASID 0
31
32 struct address_space;
33 struct mem_cgroup;
34
35 /*
36 * Each physical page in the system has a struct page associated with
37 * it to keep track of whatever it is we are using the page for at the
38 * moment. Note that we have no way to track which tasks are using
39 * a page, though if it is a pagecache page, rmap structures can tell us
40 * who is mapping it.
41 *
42 * If you allocate the page using alloc_pages(), you can use some of the
43 * space in struct page for your own purposes. The five words in the main
44 * union are available, except for bit 0 of the first word which must be
45 * kept clear. Many users use this word to store a pointer to an object
46 * which is guaranteed to be aligned. If you use the same storage as
47 * page->mapping, you must restore it to NULL before freeing the page.
48 *
49 * The mapcount field must not be used for own purposes.
50 *
51 * If you want to use the refcount field, it must be used in such a way
52 * that other CPUs temporarily incrementing and then decrementing the
53 * refcount does not cause problems. On receiving the page from
54 * alloc_pages(), the refcount will be positive.
55 *
56 * If you allocate pages of order > 0, you can use some of the fields
57 * in each subpage, but you may need to restore some of their values
58 * afterwards.
59 *
60 * SLUB uses cmpxchg_double() to atomically update its freelist and counters.
61 * That requires that freelist & counters in struct slab be adjacent and
62 * double-word aligned. Because struct slab currently just reinterprets the
63 * bits of struct page, we align all struct pages to double-word boundaries,
64 * and ensure that 'freelist' is aligned within struct slab.
65 */
66 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
67 #define _struct_page_alignment __aligned(2 * sizeof(unsigned long))
68 #else
69 #define _struct_page_alignment __aligned(sizeof(unsigned long))
70 #endif
71
72 struct page {
73 unsigned long flags; /* Atomic flags, some possibly
74 * updated asynchronously */
75 /*
76 * Five words (20/40 bytes) are available in this union.
77 * WARNING: bit 0 of the first word is used for PageTail(). That
78 * means the other users of this union MUST NOT use the bit to
79 * avoid collision and false-positive PageTail().
80 */
81 union {
82 struct { /* Page cache and anonymous pages */
83 /**
84 * @lru: Pageout list, eg. active_list protected by
85 * lruvec->lru_lock. Sometimes used as a generic list
86 * by the page owner.
87 */
88 union {
89 struct list_head lru;
90
91 /* Or, for the Unevictable "LRU list" slot */
92 struct {
93 /* Always even, to negate PageTail */
94 void *__filler;
95 /* Count page's or folio's mlocks */
96 unsigned int mlock_count;
97 };
98
99 /* Or, free page */
100 struct list_head buddy_list;
101 struct list_head pcp_list;
102 };
103 /* See page-flags.h for PAGE_MAPPING_FLAGS */
104 struct address_space *mapping;
105 union {
106 pgoff_t index; /* Our offset within mapping. */
107 unsigned long share; /* share count for fsdax */
108 };
109 /**
110 * @private: Mapping-private opaque data.
111 * Usually used for buffer_heads if PagePrivate.
112 * Used for swp_entry_t if swapcache flag set.
113 * Indicates order in the buddy system if PageBuddy.
114 */
115 unsigned long private;
116 };
117 struct { /* page_pool used by netstack */
118 /**
119 * @pp_magic: magic value to avoid recycling non
120 * page_pool allocated pages.
121 */
122 unsigned long pp_magic;
123 struct page_pool *pp;
124 unsigned long _pp_mapping_pad;
125 unsigned long dma_addr;
126 atomic_long_t pp_ref_count;
127 };
128 struct { /* Tail pages of compound page */
129 unsigned long compound_head; /* Bit zero is set */
130 };
131 struct { /* ZONE_DEVICE pages */
132 /** @pgmap: Points to the hosting device page map. */
133 struct dev_pagemap *pgmap;
134 void *zone_device_data;
135 /*
136 * ZONE_DEVICE private pages are counted as being
137 * mapped so the next 3 words hold the mapping, index,
138 * and private fields from the source anonymous or
139 * page cache page while the page is migrated to device
140 * private memory.
141 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
142 * use the mapping, index, and private fields when
143 * pmem backed DAX files are mapped.
144 */
145 };
146
147 /** @rcu_head: You can use this to free a page by RCU. */
148 struct rcu_head rcu_head;
149 };
150
151 union { /* This union is 4 bytes in size. */
152 /*
153 * For head pages of typed folios, the value stored here
154 * allows for determining what this page is used for. The
155 * tail pages of typed folios will not store a type
156 * (page_type == _mapcount == -1).
157 *
158 * See page-flags.h for a list of page types which are currently
159 * stored here.
160 *
161 * Owners of typed folios may reuse the lower 16 bit of the
162 * head page page_type field after setting the page type,
163 * but must reset these 16 bit to -1 before clearing the
164 * page type.
165 */
166 unsigned int page_type;
167
168 /*
169 * For pages that are part of non-typed folios for which mappings
170 * are tracked via the RMAP, encodes the number of times this page
171 * is directly referenced by a page table.
172 *
173 * Note that the mapcount is always initialized to -1, so that
174 * transitions both from it and to it can be tracked, using
175 * atomic_inc_and_test() and atomic_add_negative(-1).
176 */
177 atomic_t _mapcount;
178 };
179
180 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
181 atomic_t _refcount;
182
183 #ifdef CONFIG_MEMCG
184 unsigned long memcg_data;
185 #elif defined(CONFIG_SLAB_OBJ_EXT)
186 unsigned long _unused_slab_obj_exts;
187 #endif
188
189 /*
190 * On machines where all RAM is mapped into kernel address space,
191 * we can simply calculate the virtual address. On machines with
192 * highmem some memory is mapped into kernel virtual memory
193 * dynamically, so we need a place to store that address.
194 * Note that this field could be 16 bits on x86 ... ;)
195 *
196 * Architectures with slow multiplication can define
197 * WANT_PAGE_VIRTUAL in asm/page.h
198 */
199 #if defined(WANT_PAGE_VIRTUAL)
200 void *virtual; /* Kernel virtual address (NULL if
201 not kmapped, ie. highmem) */
202 #endif /* WANT_PAGE_VIRTUAL */
203
204 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
205 int _last_cpupid;
206 #endif
207
208 #ifdef CONFIG_KMSAN
209 /*
210 * KMSAN metadata for this page:
211 * - shadow page: every bit indicates whether the corresponding
212 * bit of the original page is initialized (0) or not (1);
213 * - origin page: every 4 bytes contain an id of the stack trace
214 * where the uninitialized value was created.
215 */
216 struct page *kmsan_shadow;
217 struct page *kmsan_origin;
218 #endif
219 } _struct_page_alignment;
220
221 /*
222 * struct encoded_page - a nonexistent type marking this pointer
223 *
224 * An 'encoded_page' pointer is a pointer to a regular 'struct page', but
225 * with the low bits of the pointer indicating extra context-dependent
226 * information. Only used in mmu_gather handling, and this acts as a type
227 * system check on that use.
228 *
229 * We only really have two guaranteed bits in general, although you could
230 * play with 'struct page' alignment (see CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
231 * for more.
232 *
233 * Use the supplied helper functions to endcode/decode the pointer and bits.
234 */
235 struct encoded_page;
236
237 #define ENCODED_PAGE_BITS 3ul
238
239 /* Perform rmap removal after we have flushed the TLB. */
240 #define ENCODED_PAGE_BIT_DELAY_RMAP 1ul
241
242 /*
243 * The next item in an encoded_page array is the "nr_pages" argument, specifying
244 * the number of consecutive pages starting from this page, that all belong to
245 * the same folio. For example, "nr_pages" corresponds to the number of folio
246 * references that must be dropped. If this bit is not set, "nr_pages" is
247 * implicitly 1.
248 */
249 #define ENCODED_PAGE_BIT_NR_PAGES_NEXT 2ul
250
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 gmap shadow pages
442 * (which are not linked into the user page tables) and x86
443 * pgds.
444 * @_pt_pad_1: Padding that aliases with page's compound head.
445 * @pmd_huge_pte: Protected by ptdesc->ptl, used for THPs.
446 * @__page_mapping: Aliases with page->mapping. Unused for page tables.
447 * @pt_index: Used for s390 gmap.
448 * @pt_mm: Used for x86 pgds.
449 * @pt_frag_refcount: For fragmented page table tracking. Powerpc only.
450 * @pt_share_count: Used for HugeTLB PMD page table share count.
451 * @_pt_pad_2: Padding to ensure proper alignment.
452 * @ptl: Lock for the page table.
453 * @__page_type: Same as page->page_type. Unused for page tables.
454 * @__page_refcount: Same as page refcount.
455 * @pt_memcg_data: Memcg data. Tracked for page tables here.
456 *
457 * This struct overlays struct page for now. Do not modify without a good
458 * understanding of the issues.
459 */
460 struct ptdesc {
461 unsigned long __page_flags;
462
463 union {
464 struct rcu_head pt_rcu_head;
465 struct list_head pt_list;
466 struct {
467 unsigned long _pt_pad_1;
468 pgtable_t pmd_huge_pte;
469 };
470 };
471 unsigned long __page_mapping;
472
473 union {
474 pgoff_t pt_index;
475 struct mm_struct *pt_mm;
476 atomic_t pt_frag_refcount;
477 #ifdef CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING
478 atomic_t pt_share_count;
479 #endif
480 };
481
482 union {
483 unsigned long _pt_pad_2;
484 #if ALLOC_SPLIT_PTLOCKS
485 spinlock_t *ptl;
486 #else
487 spinlock_t ptl;
488 #endif
489 };
490 unsigned int __page_type;
491 atomic_t __page_refcount;
492 #ifdef CONFIG_MEMCG
493 unsigned long pt_memcg_data;
494 #endif
495 };
496
497 #define TABLE_MATCH(pg, pt) \
498 static_assert(offsetof(struct page, pg) == offsetof(struct ptdesc, pt))
499 TABLE_MATCH(flags, __page_flags);
500 TABLE_MATCH(compound_head, pt_list);
501 TABLE_MATCH(compound_head, _pt_pad_1);
502 TABLE_MATCH(mapping, __page_mapping);
503 TABLE_MATCH(index, pt_index);
504 TABLE_MATCH(rcu_head, pt_rcu_head);
505 TABLE_MATCH(page_type, __page_type);
506 TABLE_MATCH(_refcount, __page_refcount);
507 #ifdef CONFIG_MEMCG
508 TABLE_MATCH(memcg_data, pt_memcg_data);
509 #endif
510 #undef TABLE_MATCH
511 static_assert(sizeof(struct ptdesc) <= sizeof(struct page));
512
513 #define ptdesc_page(pt) (_Generic((pt), \
514 const struct ptdesc *: (const struct page *)(pt), \
515 struct ptdesc *: (struct page *)(pt)))
516
517 #define ptdesc_folio(pt) (_Generic((pt), \
518 const struct ptdesc *: (const struct folio *)(pt), \
519 struct ptdesc *: (struct folio *)(pt)))
520
521 #define page_ptdesc(p) (_Generic((p), \
522 const struct page *: (const struct ptdesc *)(p), \
523 struct page *: (struct ptdesc *)(p)))
524
525 #ifdef CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING
ptdesc_pmd_pts_init(struct ptdesc * ptdesc)526 static inline void ptdesc_pmd_pts_init(struct ptdesc *ptdesc)
527 {
528 atomic_set(&ptdesc->pt_share_count, 0);
529 }
530
ptdesc_pmd_pts_inc(struct ptdesc * ptdesc)531 static inline void ptdesc_pmd_pts_inc(struct ptdesc *ptdesc)
532 {
533 atomic_inc(&ptdesc->pt_share_count);
534 }
535
ptdesc_pmd_pts_dec(struct ptdesc * ptdesc)536 static inline void ptdesc_pmd_pts_dec(struct ptdesc *ptdesc)
537 {
538 atomic_dec(&ptdesc->pt_share_count);
539 }
540
ptdesc_pmd_pts_count(struct ptdesc * ptdesc)541 static inline int ptdesc_pmd_pts_count(struct ptdesc *ptdesc)
542 {
543 return atomic_read(&ptdesc->pt_share_count);
544 }
545 #else
ptdesc_pmd_pts_init(struct ptdesc * ptdesc)546 static inline void ptdesc_pmd_pts_init(struct ptdesc *ptdesc)
547 {
548 }
549 #endif
550
551 /*
552 * Used for sizing the vmemmap region on some architectures
553 */
554 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
555
556 /*
557 * page_private can be used on tail pages. However, PagePrivate is only
558 * checked by the VM on the head page. So page_private on the tail pages
559 * should be used for data that's ancillary to the head page (eg attaching
560 * buffer heads to tail pages after attaching buffer heads to the head page)
561 */
562 #define page_private(page) ((page)->private)
563
set_page_private(struct page * page,unsigned long private)564 static inline void set_page_private(struct page *page, unsigned long private)
565 {
566 page->private = private;
567 }
568
folio_get_private(struct folio * folio)569 static inline void *folio_get_private(struct folio *folio)
570 {
571 return folio->private;
572 }
573
574 typedef unsigned long vm_flags_t;
575
576 /*
577 * A region containing a mapping of a non-memory backed file under NOMMU
578 * conditions. These are held in a global tree and are pinned by the VMAs that
579 * map parts of them.
580 */
581 struct vm_region {
582 struct rb_node vm_rb; /* link in global region tree */
583 vm_flags_t vm_flags; /* VMA vm_flags */
584 unsigned long vm_start; /* start address of region */
585 unsigned long vm_end; /* region initialised to here */
586 unsigned long vm_top; /* region allocated to here */
587 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
588 struct file *vm_file; /* the backing file or NULL */
589
590 int vm_usage; /* region usage count (access under nommu_region_sem) */
591 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
592 * this region */
593 };
594
595 #ifdef CONFIG_USERFAULTFD
596 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
597 struct vm_userfaultfd_ctx {
598 struct userfaultfd_ctx *ctx;
599 };
600 #else /* CONFIG_USERFAULTFD */
601 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
602 struct vm_userfaultfd_ctx {};
603 #endif /* CONFIG_USERFAULTFD */
604
605 struct anon_vma_name {
606 struct kref kref;
607 /* The name needs to be at the end because it is dynamically sized. */
608 char name[];
609 };
610
611 #ifdef CONFIG_ANON_VMA_NAME
612 /*
613 * mmap_lock should be read-locked when calling anon_vma_name(). Caller should
614 * either keep holding the lock while using the returned pointer or it should
615 * raise anon_vma_name refcount before releasing the lock.
616 */
617 struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma);
618 struct anon_vma_name *anon_vma_name_alloc(const char *name);
619 void anon_vma_name_free(struct kref *kref);
620 #else /* CONFIG_ANON_VMA_NAME */
anon_vma_name(struct vm_area_struct * vma)621 static inline struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma)
622 {
623 return NULL;
624 }
625
anon_vma_name_alloc(const char * name)626 static inline struct anon_vma_name *anon_vma_name_alloc(const char *name)
627 {
628 return NULL;
629 }
630 #endif
631
632 struct vma_lock {
633 struct rw_semaphore lock;
634 };
635
636 struct vma_numab_state {
637 /*
638 * Initialised as time in 'jiffies' after which VMA
639 * should be scanned. Delays first scan of new VMA by at
640 * least sysctl_numa_balancing_scan_delay:
641 */
642 unsigned long next_scan;
643
644 /*
645 * Time in jiffies when pids_active[] is reset to
646 * detect phase change behaviour:
647 */
648 unsigned long pids_active_reset;
649
650 /*
651 * Approximate tracking of PIDs that trapped a NUMA hinting
652 * fault. May produce false positives due to hash collisions.
653 *
654 * [0] Previous PID tracking
655 * [1] Current PID tracking
656 *
657 * Window moves after next_pid_reset has expired approximately
658 * every VMA_PID_RESET_PERIOD jiffies:
659 */
660 unsigned long pids_active[2];
661
662 /* MM scan sequence ID when scan first started after VMA creation */
663 int start_scan_seq;
664
665 /*
666 * MM scan sequence ID when the VMA was last completely scanned.
667 * A VMA is not eligible for scanning if prev_scan_seq == numa_scan_seq
668 */
669 int prev_scan_seq;
670 };
671
672 /*
673 * This struct describes a virtual memory area. There is one of these
674 * per VM-area/task. A VM area is any part of the process virtual memory
675 * space that has a special rule for the page-fault handlers (ie a shared
676 * library, the executable area etc).
677 *
678 * Only explicitly marked struct members may be accessed by RCU readers before
679 * getting a stable reference.
680 */
681 struct vm_area_struct {
682 /* The first cache line has the info for VMA tree walking. */
683
684 union {
685 struct {
686 /* VMA covers [vm_start; vm_end) addresses within mm */
687 unsigned long vm_start;
688 unsigned long vm_end;
689 };
690 #ifdef CONFIG_PER_VMA_LOCK
691 struct rcu_head vm_rcu; /* Used for deferred freeing. */
692 #endif
693 };
694
695 /*
696 * The address space we belong to.
697 * Unstable RCU readers are allowed to read this.
698 */
699 struct mm_struct *vm_mm;
700 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
701
702 /*
703 * Flags, see mm.h.
704 * To modify use vm_flags_{init|reset|set|clear|mod} functions.
705 */
706 union {
707 const vm_flags_t vm_flags;
708 vm_flags_t __private __vm_flags;
709 };
710
711 #ifdef CONFIG_PER_VMA_LOCK
712 /*
713 * Flag to indicate areas detached from the mm->mm_mt tree.
714 * Unstable RCU readers are allowed to read this.
715 */
716 bool detached;
717
718 /*
719 * Can only be written (using WRITE_ONCE()) while holding both:
720 * - mmap_lock (in write mode)
721 * - vm_lock->lock (in write mode)
722 * Can be read reliably while holding one of:
723 * - mmap_lock (in read or write mode)
724 * - vm_lock->lock (in read or write mode)
725 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
726 * while holding nothing (except RCU to keep the VMA struct allocated).
727 *
728 * This sequence counter is explicitly allowed to overflow; sequence
729 * counter reuse can only lead to occasional unnecessary use of the
730 * slowpath.
731 */
732 unsigned int vm_lock_seq;
733 /* Unstable RCU readers are allowed to read this. */
734 struct vma_lock *vm_lock;
735 #endif
736
737 /*
738 * For areas with an address space and backing store,
739 * linkage into the address_space->i_mmap interval tree.
740 *
741 */
742 struct {
743 struct rb_node rb;
744 unsigned long rb_subtree_last;
745 } shared;
746
747 /*
748 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
749 * list, after a COW of one of the file pages. A MAP_SHARED vma
750 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
751 * or brk vma (with NULL file) can only be in an anon_vma list.
752 */
753 struct list_head anon_vma_chain; /* Serialized by mmap_lock &
754 * page_table_lock */
755 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
756
757 /* Function pointers to deal with this struct. */
758 const struct vm_operations_struct *vm_ops;
759
760 /* Information about our backing store: */
761 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
762 units */
763 struct file * vm_file; /* File we map to (can be NULL). */
764 void * vm_private_data; /* was vm_pte (shared mem) */
765
766 #ifdef CONFIG_ANON_VMA_NAME
767 /*
768 * For private and shared anonymous mappings, a pointer to a null
769 * terminated string containing the name given to the vma, or NULL if
770 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access.
771 */
772 struct anon_vma_name *anon_name;
773 #endif
774 #ifdef CONFIG_SWAP
775 atomic_long_t swap_readahead_info;
776 #endif
777 #ifndef CONFIG_MMU
778 struct vm_region *vm_region; /* NOMMU mapping region */
779 #endif
780 #ifdef CONFIG_NUMA
781 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
782 #endif
783 #ifdef CONFIG_NUMA_BALANCING
784 struct vma_numab_state *numab_state; /* NUMA Balancing state */
785 #endif
786 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
787 } __randomize_layout;
788
789 #ifdef CONFIG_NUMA
790 #define vma_policy(vma) ((vma)->vm_policy)
791 #else
792 #define vma_policy(vma) NULL
793 #endif
794
795 #ifdef CONFIG_SCHED_MM_CID
796 struct mm_cid {
797 u64 time;
798 int cid;
799 int recent_cid;
800 };
801 #endif
802
803 struct kioctx_table;
804 struct iommu_mm_data;
805 struct mm_struct {
806 struct {
807 /*
808 * Fields which are often written to are placed in a separate
809 * cache line.
810 */
811 struct {
812 /**
813 * @mm_count: The number of references to &struct
814 * mm_struct (@mm_users count as 1).
815 *
816 * Use mmgrab()/mmdrop() to modify. When this drops to
817 * 0, the &struct mm_struct is freed.
818 */
819 atomic_t mm_count;
820 } ____cacheline_aligned_in_smp;
821
822 struct maple_tree mm_mt;
823
824 unsigned long mmap_base; /* base of mmap area */
825 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
826 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
827 /* Base addresses for compatible mmap() */
828 unsigned long mmap_compat_base;
829 unsigned long mmap_compat_legacy_base;
830 #endif
831 unsigned long task_size; /* size of task vm space */
832 pgd_t * pgd;
833
834 #ifdef CONFIG_MEMBARRIER
835 /**
836 * @membarrier_state: Flags controlling membarrier behavior.
837 *
838 * This field is close to @pgd to hopefully fit in the same
839 * cache-line, which needs to be touched by switch_mm().
840 */
841 atomic_t membarrier_state;
842 #endif
843
844 /**
845 * @mm_users: The number of users including userspace.
846 *
847 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
848 * drops to 0 (i.e. when the task exits and there are no other
849 * temporary reference holders), we also release a reference on
850 * @mm_count (which may then free the &struct mm_struct if
851 * @mm_count also drops to 0).
852 */
853 atomic_t mm_users;
854
855 #ifdef CONFIG_SCHED_MM_CID
856 /**
857 * @pcpu_cid: Per-cpu current cid.
858 *
859 * Keep track of the currently allocated mm_cid for each cpu.
860 * The per-cpu mm_cid values are serialized by their respective
861 * runqueue locks.
862 */
863 struct mm_cid __percpu *pcpu_cid;
864 /*
865 * @mm_cid_next_scan: Next mm_cid scan (in jiffies).
866 *
867 * When the next mm_cid scan is due (in jiffies).
868 */
869 unsigned long mm_cid_next_scan;
870 /**
871 * @nr_cpus_allowed: Number of CPUs allowed for mm.
872 *
873 * Number of CPUs allowed in the union of all mm's
874 * threads allowed CPUs.
875 */
876 unsigned int nr_cpus_allowed;
877 /**
878 * @max_nr_cid: Maximum number of allowed concurrency
879 * IDs allocated.
880 *
881 * Track the highest number of allowed concurrency IDs
882 * allocated for the mm.
883 */
884 atomic_t max_nr_cid;
885 /**
886 * @cpus_allowed_lock: Lock protecting mm cpus_allowed.
887 *
888 * Provide mutual exclusion for mm cpus_allowed and
889 * mm nr_cpus_allowed updates.
890 */
891 raw_spinlock_t cpus_allowed_lock;
892 #endif
893 #ifdef CONFIG_MMU
894 atomic_long_t pgtables_bytes; /* size of all page tables */
895 #endif
896 int map_count; /* number of VMAs */
897
898 spinlock_t page_table_lock; /* Protects page tables and some
899 * counters
900 */
901 /*
902 * With some kernel config, the current mmap_lock's offset
903 * inside 'mm_struct' is at 0x120, which is very optimal, as
904 * its two hot fields 'count' and 'owner' sit in 2 different
905 * cachelines, and when mmap_lock is highly contended, both
906 * of the 2 fields will be accessed frequently, current layout
907 * will help to reduce cache bouncing.
908 *
909 * So please be careful with adding new fields before
910 * mmap_lock, which can easily push the 2 fields into one
911 * cacheline.
912 */
913 struct rw_semaphore mmap_lock;
914
915 struct list_head mmlist; /* List of maybe swapped mm's. These
916 * are globally strung together off
917 * init_mm.mmlist, and are protected
918 * by mmlist_lock
919 */
920 #ifdef CONFIG_PER_VMA_LOCK
921 /*
922 * This field has lock-like semantics, meaning it is sometimes
923 * accessed with ACQUIRE/RELEASE semantics.
924 * Roughly speaking, incrementing the sequence number is
925 * equivalent to releasing locks on VMAs; reading the sequence
926 * number can be part of taking a read lock on a VMA.
927 * Incremented every time mmap_lock is write-locked/unlocked.
928 * Initialized to 0, therefore odd values indicate mmap_lock
929 * is write-locked and even values that it's released.
930 *
931 * Can be modified under write mmap_lock using RELEASE
932 * semantics.
933 * Can be read with no other protection when holding write
934 * mmap_lock.
935 * Can be read with ACQUIRE semantics if not holding write
936 * mmap_lock.
937 */
938 seqcount_t mm_lock_seq;
939 #endif
940
941
942 unsigned long hiwater_rss; /* High-watermark of RSS usage */
943 unsigned long hiwater_vm; /* High-water virtual memory usage */
944
945 unsigned long total_vm; /* Total pages mapped */
946 unsigned long locked_vm; /* Pages that have PG_mlocked set */
947 atomic64_t pinned_vm; /* Refcount permanently increased */
948 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
949 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
950 unsigned long stack_vm; /* VM_STACK */
951 unsigned long def_flags;
952
953 /**
954 * @write_protect_seq: Locked when any thread is write
955 * protecting pages mapped by this mm to enforce a later COW,
956 * for instance during page table copying for fork().
957 */
958 seqcount_t write_protect_seq;
959
960 spinlock_t arg_lock; /* protect the below fields */
961
962 unsigned long start_code, end_code, start_data, end_data;
963 unsigned long start_brk, brk, start_stack;
964 unsigned long arg_start, arg_end, env_start, env_end;
965
966 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
967
968 struct percpu_counter rss_stat[NR_MM_COUNTERS];
969
970 struct linux_binfmt *binfmt;
971
972 /* Architecture-specific MM context */
973 mm_context_t context;
974
975 unsigned long flags; /* Must use atomic bitops to access */
976
977 #ifdef CONFIG_AIO
978 spinlock_t ioctx_lock;
979 struct kioctx_table __rcu *ioctx_table;
980 #endif
981 #ifdef CONFIG_MEMCG
982 /*
983 * "owner" points to a task that is regarded as the canonical
984 * user/owner of this mm. All of the following must be true in
985 * order for it to be changed:
986 *
987 * current == mm->owner
988 * current->mm != mm
989 * new_owner->mm == mm
990 * new_owner->alloc_lock is held
991 */
992 struct task_struct __rcu *owner;
993 #endif
994 struct user_namespace *user_ns;
995
996 /* store ref to file /proc/<pid>/exe symlink points to */
997 struct file __rcu *exe_file;
998 #ifdef CONFIG_MMU_NOTIFIER
999 struct mmu_notifier_subscriptions *notifier_subscriptions;
1000 #endif
1001 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
1002 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
1003 #endif
1004 #ifdef CONFIG_NUMA_BALANCING
1005 /*
1006 * numa_next_scan is the next time that PTEs will be remapped
1007 * PROT_NONE to trigger NUMA hinting faults; such faults gather
1008 * statistics and migrate pages to new nodes if necessary.
1009 */
1010 unsigned long numa_next_scan;
1011
1012 /* Restart point for scanning and remapping PTEs. */
1013 unsigned long numa_scan_offset;
1014
1015 /* numa_scan_seq prevents two threads remapping PTEs. */
1016 int numa_scan_seq;
1017 #endif
1018 /*
1019 * An operation with batched TLB flushing is going on. Anything
1020 * that can move process memory needs to flush the TLB when
1021 * moving a PROT_NONE mapped page.
1022 */
1023 atomic_t tlb_flush_pending;
1024 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
1025 /* See flush_tlb_batched_pending() */
1026 atomic_t tlb_flush_batched;
1027 #endif
1028 struct uprobes_state uprobes_state;
1029 #ifdef CONFIG_PREEMPT_RT
1030 struct rcu_head delayed_drop;
1031 #endif
1032 #ifdef CONFIG_HUGETLB_PAGE
1033 atomic_long_t hugetlb_usage;
1034 #endif
1035 struct work_struct async_put_work;
1036
1037 #ifdef CONFIG_IOMMU_MM_DATA
1038 struct iommu_mm_data *iommu_mm;
1039 #endif
1040 #ifdef CONFIG_KSM
1041 /*
1042 * Represent how many pages of this process are involved in KSM
1043 * merging (not including ksm_zero_pages).
1044 */
1045 unsigned long ksm_merging_pages;
1046 /*
1047 * Represent how many pages are checked for ksm merging
1048 * including merged and not merged.
1049 */
1050 unsigned long ksm_rmap_items;
1051 /*
1052 * Represent how many empty pages are merged with kernel zero
1053 * pages when enabling KSM use_zero_pages.
1054 */
1055 atomic_long_t ksm_zero_pages;
1056 #endif /* CONFIG_KSM */
1057 #ifdef CONFIG_LRU_GEN_WALKS_MMU
1058 struct {
1059 /* this mm_struct is on lru_gen_mm_list */
1060 struct list_head list;
1061 /*
1062 * Set when switching to this mm_struct, as a hint of
1063 * whether it has been used since the last time per-node
1064 * page table walkers cleared the corresponding bits.
1065 */
1066 unsigned long bitmap;
1067 #ifdef CONFIG_MEMCG
1068 /* points to the memcg of "owner" above */
1069 struct mem_cgroup *memcg;
1070 #endif
1071 } lru_gen;
1072 #endif /* CONFIG_LRU_GEN_WALKS_MMU */
1073 } __randomize_layout;
1074
1075 /*
1076 * The mm_cpumask needs to be at the end of mm_struct, because it
1077 * is dynamically sized based on nr_cpu_ids.
1078 */
1079 unsigned long cpu_bitmap[];
1080 };
1081
1082 #define MM_MT_FLAGS (MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN | \
1083 MT_FLAGS_USE_RCU)
1084 extern struct mm_struct init_mm;
1085
1086 /* Pointer magic because the dynamic array size confuses some compilers. */
mm_init_cpumask(struct mm_struct * mm)1087 static inline void mm_init_cpumask(struct mm_struct *mm)
1088 {
1089 unsigned long cpu_bitmap = (unsigned long)mm;
1090
1091 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
1092 cpumask_clear((struct cpumask *)cpu_bitmap);
1093 }
1094
1095 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
mm_cpumask(struct mm_struct * mm)1096 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
1097 {
1098 return (struct cpumask *)&mm->cpu_bitmap;
1099 }
1100
1101 #ifdef CONFIG_LRU_GEN
1102
1103 struct lru_gen_mm_list {
1104 /* mm_struct list for page table walkers */
1105 struct list_head fifo;
1106 /* protects the list above */
1107 spinlock_t lock;
1108 };
1109
1110 #endif /* CONFIG_LRU_GEN */
1111
1112 #ifdef CONFIG_LRU_GEN_WALKS_MMU
1113
1114 void lru_gen_add_mm(struct mm_struct *mm);
1115 void lru_gen_del_mm(struct mm_struct *mm);
1116 void lru_gen_migrate_mm(struct mm_struct *mm);
1117
lru_gen_init_mm(struct mm_struct * mm)1118 static inline void lru_gen_init_mm(struct mm_struct *mm)
1119 {
1120 INIT_LIST_HEAD(&mm->lru_gen.list);
1121 mm->lru_gen.bitmap = 0;
1122 #ifdef CONFIG_MEMCG
1123 mm->lru_gen.memcg = NULL;
1124 #endif
1125 }
1126
lru_gen_use_mm(struct mm_struct * mm)1127 static inline void lru_gen_use_mm(struct mm_struct *mm)
1128 {
1129 /*
1130 * When the bitmap is set, page reclaim knows this mm_struct has been
1131 * used since the last time it cleared the bitmap. So it might be worth
1132 * walking the page tables of this mm_struct to clear the accessed bit.
1133 */
1134 WRITE_ONCE(mm->lru_gen.bitmap, -1);
1135 }
1136
1137 #else /* !CONFIG_LRU_GEN_WALKS_MMU */
1138
lru_gen_add_mm(struct mm_struct * mm)1139 static inline void lru_gen_add_mm(struct mm_struct *mm)
1140 {
1141 }
1142
lru_gen_del_mm(struct mm_struct * mm)1143 static inline void lru_gen_del_mm(struct mm_struct *mm)
1144 {
1145 }
1146
lru_gen_migrate_mm(struct mm_struct * mm)1147 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
1148 {
1149 }
1150
lru_gen_init_mm(struct mm_struct * mm)1151 static inline void lru_gen_init_mm(struct mm_struct *mm)
1152 {
1153 }
1154
lru_gen_use_mm(struct mm_struct * mm)1155 static inline void lru_gen_use_mm(struct mm_struct *mm)
1156 {
1157 }
1158
1159 #endif /* CONFIG_LRU_GEN_WALKS_MMU */
1160
1161 struct vma_iterator {
1162 struct ma_state mas;
1163 };
1164
1165 #define VMA_ITERATOR(name, __mm, __addr) \
1166 struct vma_iterator name = { \
1167 .mas = { \
1168 .tree = &(__mm)->mm_mt, \
1169 .index = __addr, \
1170 .node = NULL, \
1171 .status = ma_start, \
1172 }, \
1173 }
1174
vma_iter_init(struct vma_iterator * vmi,struct mm_struct * mm,unsigned long addr)1175 static inline void vma_iter_init(struct vma_iterator *vmi,
1176 struct mm_struct *mm, unsigned long addr)
1177 {
1178 mas_init(&vmi->mas, &mm->mm_mt, addr);
1179 }
1180
1181 #ifdef CONFIG_SCHED_MM_CID
1182
1183 enum mm_cid_state {
1184 MM_CID_UNSET = -1U, /* Unset state has lazy_put flag set. */
1185 MM_CID_LAZY_PUT = (1U << 31),
1186 };
1187
mm_cid_is_unset(int cid)1188 static inline bool mm_cid_is_unset(int cid)
1189 {
1190 return cid == MM_CID_UNSET;
1191 }
1192
mm_cid_is_lazy_put(int cid)1193 static inline bool mm_cid_is_lazy_put(int cid)
1194 {
1195 return !mm_cid_is_unset(cid) && (cid & MM_CID_LAZY_PUT);
1196 }
1197
mm_cid_is_valid(int cid)1198 static inline bool mm_cid_is_valid(int cid)
1199 {
1200 return !(cid & MM_CID_LAZY_PUT);
1201 }
1202
mm_cid_set_lazy_put(int cid)1203 static inline int mm_cid_set_lazy_put(int cid)
1204 {
1205 return cid | MM_CID_LAZY_PUT;
1206 }
1207
mm_cid_clear_lazy_put(int cid)1208 static inline int mm_cid_clear_lazy_put(int cid)
1209 {
1210 return cid & ~MM_CID_LAZY_PUT;
1211 }
1212
1213 /*
1214 * mm_cpus_allowed: Union of all mm's threads allowed CPUs.
1215 */
mm_cpus_allowed(struct mm_struct * mm)1216 static inline cpumask_t *mm_cpus_allowed(struct mm_struct *mm)
1217 {
1218 unsigned long bitmap = (unsigned long)mm;
1219
1220 bitmap += offsetof(struct mm_struct, cpu_bitmap);
1221 /* Skip cpu_bitmap */
1222 bitmap += cpumask_size();
1223 return (struct cpumask *)bitmap;
1224 }
1225
1226 /* Accessor for struct mm_struct's cidmask. */
mm_cidmask(struct mm_struct * mm)1227 static inline cpumask_t *mm_cidmask(struct mm_struct *mm)
1228 {
1229 unsigned long cid_bitmap = (unsigned long)mm_cpus_allowed(mm);
1230
1231 /* Skip mm_cpus_allowed */
1232 cid_bitmap += cpumask_size();
1233 return (struct cpumask *)cid_bitmap;
1234 }
1235
mm_init_cid(struct mm_struct * mm,struct task_struct * p)1236 static inline void mm_init_cid(struct mm_struct *mm, struct task_struct *p)
1237 {
1238 int i;
1239
1240 for_each_possible_cpu(i) {
1241 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, i);
1242
1243 pcpu_cid->cid = MM_CID_UNSET;
1244 pcpu_cid->recent_cid = MM_CID_UNSET;
1245 pcpu_cid->time = 0;
1246 }
1247 mm->nr_cpus_allowed = p->nr_cpus_allowed;
1248 atomic_set(&mm->max_nr_cid, 0);
1249 raw_spin_lock_init(&mm->cpus_allowed_lock);
1250 cpumask_copy(mm_cpus_allowed(mm), &p->cpus_mask);
1251 cpumask_clear(mm_cidmask(mm));
1252 }
1253
mm_alloc_cid_noprof(struct mm_struct * mm,struct task_struct * p)1254 static inline int mm_alloc_cid_noprof(struct mm_struct *mm, struct task_struct *p)
1255 {
1256 mm->pcpu_cid = alloc_percpu_noprof(struct mm_cid);
1257 if (!mm->pcpu_cid)
1258 return -ENOMEM;
1259 mm_init_cid(mm, p);
1260 return 0;
1261 }
1262 #define mm_alloc_cid(...) alloc_hooks(mm_alloc_cid_noprof(__VA_ARGS__))
1263
mm_destroy_cid(struct mm_struct * mm)1264 static inline void mm_destroy_cid(struct mm_struct *mm)
1265 {
1266 free_percpu(mm->pcpu_cid);
1267 mm->pcpu_cid = NULL;
1268 }
1269
mm_cid_size(void)1270 static inline unsigned int mm_cid_size(void)
1271 {
1272 return 2 * cpumask_size(); /* mm_cpus_allowed(), mm_cidmask(). */
1273 }
1274
mm_set_cpus_allowed(struct mm_struct * mm,const struct cpumask * cpumask)1275 static inline void mm_set_cpus_allowed(struct mm_struct *mm, const struct cpumask *cpumask)
1276 {
1277 struct cpumask *mm_allowed = mm_cpus_allowed(mm);
1278
1279 if (!mm)
1280 return;
1281 /* The mm_cpus_allowed is the union of each thread allowed CPUs masks. */
1282 raw_spin_lock(&mm->cpus_allowed_lock);
1283 cpumask_or(mm_allowed, mm_allowed, cpumask);
1284 WRITE_ONCE(mm->nr_cpus_allowed, cpumask_weight(mm_allowed));
1285 raw_spin_unlock(&mm->cpus_allowed_lock);
1286 }
1287 #else /* CONFIG_SCHED_MM_CID */
mm_init_cid(struct mm_struct * mm,struct task_struct * p)1288 static inline void mm_init_cid(struct mm_struct *mm, struct task_struct *p) { }
mm_alloc_cid(struct mm_struct * mm,struct task_struct * p)1289 static inline int mm_alloc_cid(struct mm_struct *mm, struct task_struct *p) { return 0; }
mm_destroy_cid(struct mm_struct * mm)1290 static inline void mm_destroy_cid(struct mm_struct *mm) { }
1291
mm_cid_size(void)1292 static inline unsigned int mm_cid_size(void)
1293 {
1294 return 0;
1295 }
mm_set_cpus_allowed(struct mm_struct * mm,const struct cpumask * cpumask)1296 static inline void mm_set_cpus_allowed(struct mm_struct *mm, const struct cpumask *cpumask) { }
1297 #endif /* CONFIG_SCHED_MM_CID */
1298
1299 struct mmu_gather;
1300 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
1301 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
1302 extern void tlb_finish_mmu(struct mmu_gather *tlb);
1303
1304 struct vm_fault;
1305
1306 /**
1307 * typedef vm_fault_t - Return type for page fault handlers.
1308 *
1309 * Page fault handlers return a bitmask of %VM_FAULT values.
1310 */
1311 typedef __bitwise unsigned int vm_fault_t;
1312
1313 /**
1314 * enum vm_fault_reason - Page fault handlers return a bitmask of
1315 * these values to tell the core VM what happened when handling the
1316 * fault. Used to decide whether a process gets delivered SIGBUS or
1317 * just gets major/minor fault counters bumped up.
1318 *
1319 * @VM_FAULT_OOM: Out Of Memory
1320 * @VM_FAULT_SIGBUS: Bad access
1321 * @VM_FAULT_MAJOR: Page read from storage
1322 * @VM_FAULT_HWPOISON: Hit poisoned small page
1323 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
1324 * in upper bits
1325 * @VM_FAULT_SIGSEGV: segmentation fault
1326 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
1327 * @VM_FAULT_LOCKED: ->fault locked the returned page
1328 * @VM_FAULT_RETRY: ->fault blocked, must retry
1329 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
1330 * @VM_FAULT_DONE_COW: ->fault has fully handled COW
1331 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
1332 * fsync() to complete (for synchronous page faults
1333 * in DAX)
1334 * @VM_FAULT_COMPLETED: ->fault completed, meanwhile mmap lock released
1335 * @VM_FAULT_HINDEX_MASK: mask HINDEX value
1336 *
1337 */
1338 enum vm_fault_reason {
1339 VM_FAULT_OOM = (__force vm_fault_t)0x000001,
1340 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
1341 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
1342 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
1343 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
1344 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
1345 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
1346 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
1347 VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
1348 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
1349 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
1350 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
1351 VM_FAULT_COMPLETED = (__force vm_fault_t)0x004000,
1352 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
1353 };
1354
1355 /* Encode hstate index for a hwpoisoned large page */
1356 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
1357 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
1358
1359 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
1360 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
1361 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
1362
1363 #define VM_FAULT_RESULT_TRACE \
1364 { VM_FAULT_OOM, "OOM" }, \
1365 { VM_FAULT_SIGBUS, "SIGBUS" }, \
1366 { VM_FAULT_MAJOR, "MAJOR" }, \
1367 { VM_FAULT_HWPOISON, "HWPOISON" }, \
1368 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
1369 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
1370 { VM_FAULT_NOPAGE, "NOPAGE" }, \
1371 { VM_FAULT_LOCKED, "LOCKED" }, \
1372 { VM_FAULT_RETRY, "RETRY" }, \
1373 { VM_FAULT_FALLBACK, "FALLBACK" }, \
1374 { VM_FAULT_DONE_COW, "DONE_COW" }, \
1375 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }, \
1376 { VM_FAULT_COMPLETED, "COMPLETED" }
1377
1378 struct vm_special_mapping {
1379 const char *name; /* The name, e.g. "[vdso]". */
1380
1381 /*
1382 * If .fault is not provided, this points to a
1383 * NULL-terminated array of pages that back the special mapping.
1384 *
1385 * This must not be NULL unless .fault is provided.
1386 */
1387 struct page **pages;
1388
1389 /*
1390 * If non-NULL, then this is called to resolve page faults
1391 * on the special mapping. If used, .pages is not checked.
1392 */
1393 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
1394 struct vm_area_struct *vma,
1395 struct vm_fault *vmf);
1396
1397 int (*mremap)(const struct vm_special_mapping *sm,
1398 struct vm_area_struct *new_vma);
1399
1400 void (*close)(const struct vm_special_mapping *sm,
1401 struct vm_area_struct *vma);
1402 };
1403
1404 enum tlb_flush_reason {
1405 TLB_FLUSH_ON_TASK_SWITCH,
1406 TLB_REMOTE_SHOOTDOWN,
1407 TLB_LOCAL_SHOOTDOWN,
1408 TLB_LOCAL_MM_SHOOTDOWN,
1409 TLB_REMOTE_SEND_IPI,
1410 TLB_REMOTE_WRONG_CPU,
1411 NR_TLB_FLUSH_REASONS,
1412 };
1413
1414 /**
1415 * enum fault_flag - Fault flag definitions.
1416 * @FAULT_FLAG_WRITE: Fault was a write fault.
1417 * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
1418 * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
1419 * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
1420 * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
1421 * @FAULT_FLAG_TRIED: The fault has been tried once.
1422 * @FAULT_FLAG_USER: The fault originated in userspace.
1423 * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
1424 * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
1425 * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
1426 * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a
1427 * COW mapping, making sure that an exclusive anon page is
1428 * mapped after the fault.
1429 * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached.
1430 * We should only access orig_pte if this flag set.
1431 * @FAULT_FLAG_VMA_LOCK: The fault is handled under VMA lock.
1432 *
1433 * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
1434 * whether we would allow page faults to retry by specifying these two
1435 * fault flags correctly. Currently there can be three legal combinations:
1436 *
1437 * (a) ALLOW_RETRY and !TRIED: this means the page fault allows retry, and
1438 * this is the first try
1439 *
1440 * (b) ALLOW_RETRY and TRIED: this means the page fault allows retry, and
1441 * we've already tried at least once
1442 *
1443 * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
1444 *
1445 * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
1446 * be used. Note that page faults can be allowed to retry for multiple times,
1447 * in which case we'll have an initial fault with flags (a) then later on
1448 * continuous faults with flags (b). We should always try to detect pending
1449 * signals before a retry to make sure the continuous page faults can still be
1450 * interrupted if necessary.
1451 *
1452 * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal.
1453 * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when
1454 * applied to mappings that are not COW mappings.
1455 */
1456 enum fault_flag {
1457 FAULT_FLAG_WRITE = 1 << 0,
1458 FAULT_FLAG_MKWRITE = 1 << 1,
1459 FAULT_FLAG_ALLOW_RETRY = 1 << 2,
1460 FAULT_FLAG_RETRY_NOWAIT = 1 << 3,
1461 FAULT_FLAG_KILLABLE = 1 << 4,
1462 FAULT_FLAG_TRIED = 1 << 5,
1463 FAULT_FLAG_USER = 1 << 6,
1464 FAULT_FLAG_REMOTE = 1 << 7,
1465 FAULT_FLAG_INSTRUCTION = 1 << 8,
1466 FAULT_FLAG_INTERRUPTIBLE = 1 << 9,
1467 FAULT_FLAG_UNSHARE = 1 << 10,
1468 FAULT_FLAG_ORIG_PTE_VALID = 1 << 11,
1469 FAULT_FLAG_VMA_LOCK = 1 << 12,
1470 };
1471
1472 typedef unsigned int __bitwise zap_flags_t;
1473
1474 /* Flags for clear_young_dirty_ptes(). */
1475 typedef int __bitwise cydp_t;
1476
1477 /* Clear the access bit */
1478 #define CYDP_CLEAR_YOUNG ((__force cydp_t)BIT(0))
1479
1480 /* Clear the dirty bit */
1481 #define CYDP_CLEAR_DIRTY ((__force cydp_t)BIT(1))
1482
1483 /*
1484 * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each
1485 * other. Here is what they mean, and how to use them:
1486 *
1487 *
1488 * FIXME: For pages which are part of a filesystem, mappings are subject to the
1489 * lifetime enforced by the filesystem and we need guarantees that longterm
1490 * users like RDMA and V4L2 only establish mappings which coordinate usage with
1491 * the filesystem. Ideas for this coordination include revoking the longterm
1492 * pin, delaying writeback, bounce buffer page writeback, etc. As FS DAX was
1493 * added after the problem with filesystems was found FS DAX VMAs are
1494 * specifically failed. Filesystem pages are still subject to bugs and use of
1495 * FOLL_LONGTERM should be avoided on those pages.
1496 *
1497 * In the CMA case: long term pins in a CMA region would unnecessarily fragment
1498 * that region. And so, CMA attempts to migrate the page before pinning, when
1499 * FOLL_LONGTERM is specified.
1500 *
1501 * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount,
1502 * but an additional pin counting system) will be invoked. This is intended for
1503 * anything that gets a page reference and then touches page data (for example,
1504 * Direct IO). This lets the filesystem know that some non-file-system entity is
1505 * potentially changing the pages' data. In contrast to FOLL_GET (whose pages
1506 * are released via put_page()), FOLL_PIN pages must be released, ultimately, by
1507 * a call to unpin_user_page().
1508 *
1509 * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different
1510 * and separate refcounting mechanisms, however, and that means that each has
1511 * its own acquire and release mechanisms:
1512 *
1513 * FOLL_GET: get_user_pages*() to acquire, and put_page() to release.
1514 *
1515 * FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release.
1516 *
1517 * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call.
1518 * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based
1519 * calls applied to them, and that's perfectly OK. This is a constraint on the
1520 * callers, not on the pages.)
1521 *
1522 * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never
1523 * directly by the caller. That's in order to help avoid mismatches when
1524 * releasing pages: get_user_pages*() pages must be released via put_page(),
1525 * while pin_user_pages*() pages must be released via unpin_user_page().
1526 *
1527 * Please see Documentation/core-api/pin_user_pages.rst for more information.
1528 */
1529
1530 enum {
1531 /* check pte is writable */
1532 FOLL_WRITE = 1 << 0,
1533 /* do get_page on page */
1534 FOLL_GET = 1 << 1,
1535 /* give error on hole if it would be zero */
1536 FOLL_DUMP = 1 << 2,
1537 /* get_user_pages read/write w/o permission */
1538 FOLL_FORCE = 1 << 3,
1539 /*
1540 * if a disk transfer is needed, start the IO and return without waiting
1541 * upon it
1542 */
1543 FOLL_NOWAIT = 1 << 4,
1544 /* do not fault in pages */
1545 FOLL_NOFAULT = 1 << 5,
1546 /* check page is hwpoisoned */
1547 FOLL_HWPOISON = 1 << 6,
1548 /* don't do file mappings */
1549 FOLL_ANON = 1 << 7,
1550 /*
1551 * FOLL_LONGTERM indicates that the page will be held for an indefinite
1552 * time period _often_ under userspace control. This is in contrast to
1553 * iov_iter_get_pages(), whose usages are transient.
1554 */
1555 FOLL_LONGTERM = 1 << 8,
1556 /* split huge pmd before returning */
1557 FOLL_SPLIT_PMD = 1 << 9,
1558 /* allow returning PCI P2PDMA pages */
1559 FOLL_PCI_P2PDMA = 1 << 10,
1560 /* allow interrupts from generic signals */
1561 FOLL_INTERRUPTIBLE = 1 << 11,
1562 /*
1563 * Always honor (trigger) NUMA hinting faults.
1564 *
1565 * FOLL_WRITE implicitly honors NUMA hinting faults because a
1566 * PROT_NONE-mapped page is not writable (exceptions with FOLL_FORCE
1567 * apply). get_user_pages_fast_only() always implicitly honors NUMA
1568 * hinting faults.
1569 */
1570 FOLL_HONOR_NUMA_FAULT = 1 << 12,
1571
1572 /* See also internal only FOLL flags in mm/internal.h */
1573 };
1574
1575 /* mm flags */
1576
1577 /*
1578 * The first two bits represent core dump modes for set-user-ID,
1579 * the modes are SUID_DUMP_* defined in linux/sched/coredump.h
1580 */
1581 #define MMF_DUMPABLE_BITS 2
1582 #define MMF_DUMPABLE_MASK ((1 << MMF_DUMPABLE_BITS) - 1)
1583 /* coredump filter bits */
1584 #define MMF_DUMP_ANON_PRIVATE 2
1585 #define MMF_DUMP_ANON_SHARED 3
1586 #define MMF_DUMP_MAPPED_PRIVATE 4
1587 #define MMF_DUMP_MAPPED_SHARED 5
1588 #define MMF_DUMP_ELF_HEADERS 6
1589 #define MMF_DUMP_HUGETLB_PRIVATE 7
1590 #define MMF_DUMP_HUGETLB_SHARED 8
1591 #define MMF_DUMP_DAX_PRIVATE 9
1592 #define MMF_DUMP_DAX_SHARED 10
1593
1594 #define MMF_DUMP_FILTER_SHIFT MMF_DUMPABLE_BITS
1595 #define MMF_DUMP_FILTER_BITS 9
1596 #define MMF_DUMP_FILTER_MASK \
1597 (((1 << MMF_DUMP_FILTER_BITS) - 1) << MMF_DUMP_FILTER_SHIFT)
1598 #define MMF_DUMP_FILTER_DEFAULT \
1599 ((1 << MMF_DUMP_ANON_PRIVATE) | (1 << MMF_DUMP_ANON_SHARED) |\
1600 (1 << MMF_DUMP_HUGETLB_PRIVATE) | MMF_DUMP_MASK_DEFAULT_ELF)
1601
1602 #ifdef CONFIG_CORE_DUMP_DEFAULT_ELF_HEADERS
1603 # define MMF_DUMP_MASK_DEFAULT_ELF (1 << MMF_DUMP_ELF_HEADERS)
1604 #else
1605 # define MMF_DUMP_MASK_DEFAULT_ELF 0
1606 #endif
1607 /* leave room for more dump flags */
1608 #define MMF_VM_MERGEABLE 16 /* KSM may merge identical pages */
1609 #define MMF_VM_HUGEPAGE 17 /* set when mm is available for khugepaged */
1610
1611 /*
1612 * This one-shot flag is dropped due to necessity of changing exe once again
1613 * on NFS restore
1614 */
1615 //#define MMF_EXE_FILE_CHANGED 18 /* see prctl_set_mm_exe_file() */
1616
1617 #define MMF_HAS_UPROBES 19 /* has uprobes */
1618 #define MMF_RECALC_UPROBES 20 /* MMF_HAS_UPROBES can be wrong */
1619 #define MMF_OOM_SKIP 21 /* mm is of no interest for the OOM killer */
1620 #define MMF_UNSTABLE 22 /* mm is unstable for copy_from_user */
1621 #define MMF_HUGE_ZERO_PAGE 23 /* mm has ever used the global huge zero page */
1622 #define MMF_DISABLE_THP 24 /* disable THP for all VMAs */
1623 #define MMF_DISABLE_THP_MASK (1 << MMF_DISABLE_THP)
1624 #define MMF_OOM_REAP_QUEUED 25 /* mm was queued for oom_reaper */
1625 #define MMF_MULTIPROCESS 26 /* mm is shared between processes */
1626 /*
1627 * MMF_HAS_PINNED: Whether this mm has pinned any pages. This can be either
1628 * replaced in the future by mm.pinned_vm when it becomes stable, or grow into
1629 * a counter on its own. We're aggresive on this bit for now: even if the
1630 * pinned pages were unpinned later on, we'll still keep this bit set for the
1631 * lifecycle of this mm, just for simplicity.
1632 */
1633 #define MMF_HAS_PINNED 27 /* FOLL_PIN has run, never cleared */
1634
1635 #define MMF_HAS_MDWE 28
1636 #define MMF_HAS_MDWE_MASK (1 << MMF_HAS_MDWE)
1637
1638
1639 #define MMF_HAS_MDWE_NO_INHERIT 29
1640
1641 #define MMF_VM_MERGE_ANY 30
1642 #define MMF_VM_MERGE_ANY_MASK (1 << MMF_VM_MERGE_ANY)
1643
1644 #define MMF_TOPDOWN 31 /* mm searches top down by default */
1645 #define MMF_TOPDOWN_MASK (1 << MMF_TOPDOWN)
1646
1647 #define MMF_INIT_MASK (MMF_DUMPABLE_MASK | MMF_DUMP_FILTER_MASK |\
1648 MMF_DISABLE_THP_MASK | MMF_HAS_MDWE_MASK |\
1649 MMF_VM_MERGE_ANY_MASK | MMF_TOPDOWN_MASK)
1650
mmf_init_flags(unsigned long flags)1651 static inline unsigned long mmf_init_flags(unsigned long flags)
1652 {
1653 if (flags & (1UL << MMF_HAS_MDWE_NO_INHERIT))
1654 flags &= ~((1UL << MMF_HAS_MDWE) |
1655 (1UL << MMF_HAS_MDWE_NO_INHERIT));
1656 return flags & MMF_INIT_MASK;
1657 }
1658
1659 #endif /* _LINUX_MM_TYPES_H */
1660