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 and x86. 442 * @_pt_pad_1: Padding that aliases with page's compound head. 443 * @pmd_huge_pte: Protected by ptdesc->ptl, used for THPs. 444 * @__page_mapping: Aliases with page->mapping. Unused for page tables. 445 * @pt_index: Used for s390 gmap. 446 * @pt_mm: Used for x86 pgds. 447 * @pt_frag_refcount: For fragmented page table tracking. Powerpc only. 448 * @_pt_pad_2: Padding to ensure proper alignment. 449 * @ptl: Lock for the page table. 450 * @__page_type: Same as page->page_type. Unused for page tables. 451 * @__page_refcount: Same as page refcount. 452 * @pt_memcg_data: Memcg data. Tracked for page tables here. 453 * 454 * This struct overlays struct page for now. Do not modify without a good 455 * understanding of the issues. 456 */ 457 struct ptdesc { 458 unsigned long __page_flags; 459 460 union { 461 struct rcu_head pt_rcu_head; 462 struct list_head pt_list; 463 struct { 464 unsigned long _pt_pad_1; 465 pgtable_t pmd_huge_pte; 466 }; 467 }; 468 unsigned long __page_mapping; 469 470 union { 471 pgoff_t pt_index; 472 struct mm_struct *pt_mm; 473 atomic_t pt_frag_refcount; 474 }; 475 476 union { 477 unsigned long _pt_pad_2; 478 #if ALLOC_SPLIT_PTLOCKS 479 spinlock_t *ptl; 480 #else 481 spinlock_t ptl; 482 #endif 483 }; 484 unsigned int __page_type; 485 atomic_t __page_refcount; 486 #ifdef CONFIG_MEMCG 487 unsigned long pt_memcg_data; 488 #endif 489 }; 490 491 #define TABLE_MATCH(pg, pt) \ 492 static_assert(offsetof(struct page, pg) == offsetof(struct ptdesc, pt)) 493 TABLE_MATCH(flags, __page_flags); 494 TABLE_MATCH(compound_head, pt_list); 495 TABLE_MATCH(compound_head, _pt_pad_1); 496 TABLE_MATCH(mapping, __page_mapping); 497 TABLE_MATCH(index, pt_index); 498 TABLE_MATCH(rcu_head, pt_rcu_head); 499 TABLE_MATCH(page_type, __page_type); 500 TABLE_MATCH(_refcount, __page_refcount); 501 #ifdef CONFIG_MEMCG 502 TABLE_MATCH(memcg_data, pt_memcg_data); 503 #endif 504 #undef TABLE_MATCH 505 static_assert(sizeof(struct ptdesc) <= sizeof(struct page)); 506 507 #define ptdesc_page(pt) (_Generic((pt), \ 508 const struct ptdesc *: (const struct page *)(pt), \ 509 struct ptdesc *: (struct page *)(pt))) 510 511 #define ptdesc_folio(pt) (_Generic((pt), \ 512 const struct ptdesc *: (const struct folio *)(pt), \ 513 struct ptdesc *: (struct folio *)(pt))) 514 515 #define page_ptdesc(p) (_Generic((p), \ 516 const struct page *: (const struct ptdesc *)(p), \ 517 struct page *: (struct ptdesc *)(p))) 518 519 /* 520 * Used for sizing the vmemmap region on some architectures 521 */ 522 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page))) 523 524 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK) 525 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE) 526 527 /* 528 * page_private can be used on tail pages. However, PagePrivate is only 529 * checked by the VM on the head page. So page_private on the tail pages 530 * should be used for data that's ancillary to the head page (eg attaching 531 * buffer heads to tail pages after attaching buffer heads to the head page) 532 */ 533 #define page_private(page) ((page)->private) 534 set_page_private(struct page * page,unsigned long private)535 static inline void set_page_private(struct page *page, unsigned long private) 536 { 537 page->private = private; 538 } 539 folio_get_private(struct folio * folio)540 static inline void *folio_get_private(struct folio *folio) 541 { 542 return folio->private; 543 } 544 545 struct page_frag_cache { 546 void * va; 547 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) 548 __u16 offset; 549 __u16 size; 550 #else 551 __u32 offset; 552 #endif 553 /* we maintain a pagecount bias, so that we dont dirty cache line 554 * containing page->_refcount every time we allocate a fragment. 555 */ 556 unsigned int pagecnt_bias; 557 bool pfmemalloc; 558 }; 559 560 typedef unsigned long vm_flags_t; 561 562 /* 563 * A region containing a mapping of a non-memory backed file under NOMMU 564 * conditions. These are held in a global tree and are pinned by the VMAs that 565 * map parts of them. 566 */ 567 struct vm_region { 568 struct rb_node vm_rb; /* link in global region tree */ 569 vm_flags_t vm_flags; /* VMA vm_flags */ 570 unsigned long vm_start; /* start address of region */ 571 unsigned long vm_end; /* region initialised to here */ 572 unsigned long vm_top; /* region allocated to here */ 573 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */ 574 struct file *vm_file; /* the backing file or NULL */ 575 576 int vm_usage; /* region usage count (access under nommu_region_sem) */ 577 bool vm_icache_flushed : 1; /* true if the icache has been flushed for 578 * this region */ 579 }; 580 581 #ifdef CONFIG_USERFAULTFD 582 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, }) 583 struct vm_userfaultfd_ctx { 584 struct userfaultfd_ctx *ctx; 585 }; 586 #else /* CONFIG_USERFAULTFD */ 587 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {}) 588 struct vm_userfaultfd_ctx {}; 589 #endif /* CONFIG_USERFAULTFD */ 590 591 struct anon_vma_name { 592 struct kref kref; 593 /* The name needs to be at the end because it is dynamically sized. */ 594 char name[]; 595 }; 596 597 #ifdef CONFIG_ANON_VMA_NAME 598 /* 599 * mmap_lock should be read-locked when calling anon_vma_name(). Caller should 600 * either keep holding the lock while using the returned pointer or it should 601 * raise anon_vma_name refcount before releasing the lock. 602 */ 603 struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma); 604 struct anon_vma_name *anon_vma_name_alloc(const char *name); 605 void anon_vma_name_free(struct kref *kref); 606 #else /* CONFIG_ANON_VMA_NAME */ anon_vma_name(struct vm_area_struct * vma)607 static inline struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma) 608 { 609 return NULL; 610 } 611 anon_vma_name_alloc(const char * name)612 static inline struct anon_vma_name *anon_vma_name_alloc(const char *name) 613 { 614 return NULL; 615 } 616 #endif 617 618 struct vma_lock { 619 struct rw_semaphore lock; 620 }; 621 622 struct vma_numab_state { 623 /* 624 * Initialised as time in 'jiffies' after which VMA 625 * should be scanned. Delays first scan of new VMA by at 626 * least sysctl_numa_balancing_scan_delay: 627 */ 628 unsigned long next_scan; 629 630 /* 631 * Time in jiffies when pids_active[] is reset to 632 * detect phase change behaviour: 633 */ 634 unsigned long pids_active_reset; 635 636 /* 637 * Approximate tracking of PIDs that trapped a NUMA hinting 638 * fault. May produce false positives due to hash collisions. 639 * 640 * [0] Previous PID tracking 641 * [1] Current PID tracking 642 * 643 * Window moves after next_pid_reset has expired approximately 644 * every VMA_PID_RESET_PERIOD jiffies: 645 */ 646 unsigned long pids_active[2]; 647 648 /* MM scan sequence ID when scan first started after VMA creation */ 649 int start_scan_seq; 650 651 /* 652 * MM scan sequence ID when the VMA was last completely scanned. 653 * A VMA is not eligible for scanning if prev_scan_seq == numa_scan_seq 654 */ 655 int prev_scan_seq; 656 }; 657 658 /* 659 * This struct describes a virtual memory area. There is one of these 660 * per VM-area/task. A VM area is any part of the process virtual memory 661 * space that has a special rule for the page-fault handlers (ie a shared 662 * library, the executable area etc). 663 * 664 * Only explicitly marked struct members may be accessed by RCU readers before 665 * getting a stable reference. 666 */ 667 struct vm_area_struct { 668 /* The first cache line has the info for VMA tree walking. */ 669 670 union { 671 struct { 672 /* VMA covers [vm_start; vm_end) addresses within mm */ 673 unsigned long vm_start; 674 unsigned long vm_end; 675 }; 676 #ifdef CONFIG_PER_VMA_LOCK 677 struct rcu_head vm_rcu; /* Used for deferred freeing. */ 678 #endif 679 }; 680 681 /* 682 * The address space we belong to. 683 * Unstable RCU readers are allowed to read this. 684 */ 685 struct mm_struct *vm_mm; 686 pgprot_t vm_page_prot; /* Access permissions of this VMA. */ 687 688 /* 689 * Flags, see mm.h. 690 * To modify use vm_flags_{init|reset|set|clear|mod} functions. 691 */ 692 union { 693 const vm_flags_t vm_flags; 694 vm_flags_t __private __vm_flags; 695 }; 696 697 #ifdef CONFIG_PER_VMA_LOCK 698 /* 699 * Flag to indicate areas detached from the mm->mm_mt tree. 700 * Unstable RCU readers are allowed to read this. 701 */ 702 bool detached; 703 704 /* 705 * Can only be written (using WRITE_ONCE()) while holding both: 706 * - mmap_lock (in write mode) 707 * - vm_lock->lock (in write mode) 708 * Can be read reliably while holding one of: 709 * - mmap_lock (in read or write mode) 710 * - vm_lock->lock (in read or write mode) 711 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout 712 * while holding nothing (except RCU to keep the VMA struct allocated). 713 * 714 * This sequence counter is explicitly allowed to overflow; sequence 715 * counter reuse can only lead to occasional unnecessary use of the 716 * slowpath. 717 */ 718 int vm_lock_seq; 719 /* Unstable RCU readers are allowed to read this. */ 720 struct vma_lock *vm_lock; 721 #endif 722 723 /* 724 * For areas with an address space and backing store, 725 * linkage into the address_space->i_mmap interval tree. 726 * 727 */ 728 struct { 729 struct rb_node rb; 730 unsigned long rb_subtree_last; 731 } shared; 732 733 /* 734 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma 735 * list, after a COW of one of the file pages. A MAP_SHARED vma 736 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack 737 * or brk vma (with NULL file) can only be in an anon_vma list. 738 */ 739 struct list_head anon_vma_chain; /* Serialized by mmap_lock & 740 * page_table_lock */ 741 struct anon_vma *anon_vma; /* Serialized by page_table_lock */ 742 743 /* Function pointers to deal with this struct. */ 744 const struct vm_operations_struct *vm_ops; 745 746 /* Information about our backing store: */ 747 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE 748 units */ 749 struct file * vm_file; /* File we map to (can be NULL). */ 750 void * vm_private_data; /* was vm_pte (shared mem) */ 751 752 #ifdef CONFIG_ANON_VMA_NAME 753 /* 754 * For private and shared anonymous mappings, a pointer to a null 755 * terminated string containing the name given to the vma, or NULL if 756 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access. 757 */ 758 struct anon_vma_name *anon_name; 759 #endif 760 #ifdef CONFIG_SWAP 761 atomic_long_t swap_readahead_info; 762 #endif 763 #ifndef CONFIG_MMU 764 struct vm_region *vm_region; /* NOMMU mapping region */ 765 #endif 766 #ifdef CONFIG_NUMA 767 struct mempolicy *vm_policy; /* NUMA policy for the VMA */ 768 #endif 769 #ifdef CONFIG_NUMA_BALANCING 770 struct vma_numab_state *numab_state; /* NUMA Balancing state */ 771 #endif 772 struct vm_userfaultfd_ctx vm_userfaultfd_ctx; 773 } __randomize_layout; 774 775 #ifdef CONFIG_NUMA 776 #define vma_policy(vma) ((vma)->vm_policy) 777 #else 778 #define vma_policy(vma) NULL 779 #endif 780 781 #ifdef CONFIG_SCHED_MM_CID 782 struct mm_cid { 783 u64 time; 784 int cid; 785 }; 786 #endif 787 788 struct kioctx_table; 789 struct iommu_mm_data; 790 struct mm_struct { 791 struct { 792 /* 793 * Fields which are often written to are placed in a separate 794 * cache line. 795 */ 796 struct { 797 /** 798 * @mm_count: The number of references to &struct 799 * mm_struct (@mm_users count as 1). 800 * 801 * Use mmgrab()/mmdrop() to modify. When this drops to 802 * 0, the &struct mm_struct is freed. 803 */ 804 atomic_t mm_count; 805 } ____cacheline_aligned_in_smp; 806 807 struct maple_tree mm_mt; 808 809 unsigned long mmap_base; /* base of mmap area */ 810 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */ 811 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES 812 /* Base addresses for compatible mmap() */ 813 unsigned long mmap_compat_base; 814 unsigned long mmap_compat_legacy_base; 815 #endif 816 unsigned long task_size; /* size of task vm space */ 817 pgd_t * pgd; 818 819 #ifdef CONFIG_MEMBARRIER 820 /** 821 * @membarrier_state: Flags controlling membarrier behavior. 822 * 823 * This field is close to @pgd to hopefully fit in the same 824 * cache-line, which needs to be touched by switch_mm(). 825 */ 826 atomic_t membarrier_state; 827 #endif 828 829 /** 830 * @mm_users: The number of users including userspace. 831 * 832 * Use mmget()/mmget_not_zero()/mmput() to modify. When this 833 * drops to 0 (i.e. when the task exits and there are no other 834 * temporary reference holders), we also release a reference on 835 * @mm_count (which may then free the &struct mm_struct if 836 * @mm_count also drops to 0). 837 */ 838 atomic_t mm_users; 839 840 #ifdef CONFIG_SCHED_MM_CID 841 /** 842 * @pcpu_cid: Per-cpu current cid. 843 * 844 * Keep track of the currently allocated mm_cid for each cpu. 845 * The per-cpu mm_cid values are serialized by their respective 846 * runqueue locks. 847 */ 848 struct mm_cid __percpu *pcpu_cid; 849 /* 850 * @mm_cid_next_scan: Next mm_cid scan (in jiffies). 851 * 852 * When the next mm_cid scan is due (in jiffies). 853 */ 854 unsigned long mm_cid_next_scan; 855 #endif 856 #ifdef CONFIG_MMU 857 atomic_long_t pgtables_bytes; /* size of all page tables */ 858 #endif 859 int map_count; /* number of VMAs */ 860 861 spinlock_t page_table_lock; /* Protects page tables and some 862 * counters 863 */ 864 /* 865 * With some kernel config, the current mmap_lock's offset 866 * inside 'mm_struct' is at 0x120, which is very optimal, as 867 * its two hot fields 'count' and 'owner' sit in 2 different 868 * cachelines, and when mmap_lock is highly contended, both 869 * of the 2 fields will be accessed frequently, current layout 870 * will help to reduce cache bouncing. 871 * 872 * So please be careful with adding new fields before 873 * mmap_lock, which can easily push the 2 fields into one 874 * cacheline. 875 */ 876 struct rw_semaphore mmap_lock; 877 878 struct list_head mmlist; /* List of maybe swapped mm's. These 879 * are globally strung together off 880 * init_mm.mmlist, and are protected 881 * by mmlist_lock 882 */ 883 #ifdef CONFIG_PER_VMA_LOCK 884 /* 885 * This field has lock-like semantics, meaning it is sometimes 886 * accessed with ACQUIRE/RELEASE semantics. 887 * Roughly speaking, incrementing the sequence number is 888 * equivalent to releasing locks on VMAs; reading the sequence 889 * number can be part of taking a read lock on a VMA. 890 * 891 * Can be modified under write mmap_lock using RELEASE 892 * semantics. 893 * Can be read with no other protection when holding write 894 * mmap_lock. 895 * Can be read with ACQUIRE semantics if not holding write 896 * mmap_lock. 897 */ 898 int mm_lock_seq; 899 #endif 900 901 902 unsigned long hiwater_rss; /* High-watermark of RSS usage */ 903 unsigned long hiwater_vm; /* High-water virtual memory usage */ 904 905 unsigned long total_vm; /* Total pages mapped */ 906 unsigned long locked_vm; /* Pages that have PG_mlocked set */ 907 atomic64_t pinned_vm; /* Refcount permanently increased */ 908 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */ 909 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */ 910 unsigned long stack_vm; /* VM_STACK */ 911 unsigned long def_flags; 912 913 /** 914 * @write_protect_seq: Locked when any thread is write 915 * protecting pages mapped by this mm to enforce a later COW, 916 * for instance during page table copying for fork(). 917 */ 918 seqcount_t write_protect_seq; 919 920 spinlock_t arg_lock; /* protect the below fields */ 921 922 unsigned long start_code, end_code, start_data, end_data; 923 unsigned long start_brk, brk, start_stack; 924 unsigned long arg_start, arg_end, env_start, env_end; 925 926 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */ 927 928 struct percpu_counter rss_stat[NR_MM_COUNTERS]; 929 930 struct linux_binfmt *binfmt; 931 932 /* Architecture-specific MM context */ 933 mm_context_t context; 934 935 unsigned long flags; /* Must use atomic bitops to access */ 936 937 #ifdef CONFIG_AIO 938 spinlock_t ioctx_lock; 939 struct kioctx_table __rcu *ioctx_table; 940 #endif 941 #ifdef CONFIG_MEMCG 942 /* 943 * "owner" points to a task that is regarded as the canonical 944 * user/owner of this mm. All of the following must be true in 945 * order for it to be changed: 946 * 947 * current == mm->owner 948 * current->mm != mm 949 * new_owner->mm == mm 950 * new_owner->alloc_lock is held 951 */ 952 struct task_struct __rcu *owner; 953 #endif 954 struct user_namespace *user_ns; 955 956 /* store ref to file /proc/<pid>/exe symlink points to */ 957 struct file __rcu *exe_file; 958 #ifdef CONFIG_MMU_NOTIFIER 959 struct mmu_notifier_subscriptions *notifier_subscriptions; 960 #endif 961 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS) 962 pgtable_t pmd_huge_pte; /* protected by page_table_lock */ 963 #endif 964 #ifdef CONFIG_NUMA_BALANCING 965 /* 966 * numa_next_scan is the next time that PTEs will be remapped 967 * PROT_NONE to trigger NUMA hinting faults; such faults gather 968 * statistics and migrate pages to new nodes if necessary. 969 */ 970 unsigned long numa_next_scan; 971 972 /* Restart point for scanning and remapping PTEs. */ 973 unsigned long numa_scan_offset; 974 975 /* numa_scan_seq prevents two threads remapping PTEs. */ 976 int numa_scan_seq; 977 #endif 978 /* 979 * An operation with batched TLB flushing is going on. Anything 980 * that can move process memory needs to flush the TLB when 981 * moving a PROT_NONE mapped page. 982 */ 983 atomic_t tlb_flush_pending; 984 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH 985 /* See flush_tlb_batched_pending() */ 986 atomic_t tlb_flush_batched; 987 #endif 988 struct uprobes_state uprobes_state; 989 #ifdef CONFIG_PREEMPT_RT 990 struct rcu_head delayed_drop; 991 #endif 992 #ifdef CONFIG_HUGETLB_PAGE 993 atomic_long_t hugetlb_usage; 994 #endif 995 struct work_struct async_put_work; 996 997 #ifdef CONFIG_IOMMU_MM_DATA 998 struct iommu_mm_data *iommu_mm; 999 #endif 1000 #ifdef CONFIG_KSM 1001 /* 1002 * Represent how many pages of this process are involved in KSM 1003 * merging (not including ksm_zero_pages). 1004 */ 1005 unsigned long ksm_merging_pages; 1006 /* 1007 * Represent how many pages are checked for ksm merging 1008 * including merged and not merged. 1009 */ 1010 unsigned long ksm_rmap_items; 1011 /* 1012 * Represent how many empty pages are merged with kernel zero 1013 * pages when enabling KSM use_zero_pages. 1014 */ 1015 atomic_long_t ksm_zero_pages; 1016 #endif /* CONFIG_KSM */ 1017 #ifdef CONFIG_LRU_GEN_WALKS_MMU 1018 struct { 1019 /* this mm_struct is on lru_gen_mm_list */ 1020 struct list_head list; 1021 /* 1022 * Set when switching to this mm_struct, as a hint of 1023 * whether it has been used since the last time per-node 1024 * page table walkers cleared the corresponding bits. 1025 */ 1026 unsigned long bitmap; 1027 #ifdef CONFIG_MEMCG 1028 /* points to the memcg of "owner" above */ 1029 struct mem_cgroup *memcg; 1030 #endif 1031 } lru_gen; 1032 #endif /* CONFIG_LRU_GEN_WALKS_MMU */ 1033 } __randomize_layout; 1034 1035 /* 1036 * The mm_cpumask needs to be at the end of mm_struct, because it 1037 * is dynamically sized based on nr_cpu_ids. 1038 */ 1039 unsigned long cpu_bitmap[]; 1040 }; 1041 1042 #define MM_MT_FLAGS (MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN | \ 1043 MT_FLAGS_USE_RCU) 1044 extern struct mm_struct init_mm; 1045 1046 /* Pointer magic because the dynamic array size confuses some compilers. */ mm_init_cpumask(struct mm_struct * mm)1047 static inline void mm_init_cpumask(struct mm_struct *mm) 1048 { 1049 unsigned long cpu_bitmap = (unsigned long)mm; 1050 1051 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap); 1052 cpumask_clear((struct cpumask *)cpu_bitmap); 1053 } 1054 1055 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */ mm_cpumask(struct mm_struct * mm)1056 static inline cpumask_t *mm_cpumask(struct mm_struct *mm) 1057 { 1058 return (struct cpumask *)&mm->cpu_bitmap; 1059 } 1060 1061 #ifdef CONFIG_LRU_GEN 1062 1063 struct lru_gen_mm_list { 1064 /* mm_struct list for page table walkers */ 1065 struct list_head fifo; 1066 /* protects the list above */ 1067 spinlock_t lock; 1068 }; 1069 1070 #endif /* CONFIG_LRU_GEN */ 1071 1072 #ifdef CONFIG_LRU_GEN_WALKS_MMU 1073 1074 void lru_gen_add_mm(struct mm_struct *mm); 1075 void lru_gen_del_mm(struct mm_struct *mm); 1076 void lru_gen_migrate_mm(struct mm_struct *mm); 1077 lru_gen_init_mm(struct mm_struct * mm)1078 static inline void lru_gen_init_mm(struct mm_struct *mm) 1079 { 1080 INIT_LIST_HEAD(&mm->lru_gen.list); 1081 mm->lru_gen.bitmap = 0; 1082 #ifdef CONFIG_MEMCG 1083 mm->lru_gen.memcg = NULL; 1084 #endif 1085 } 1086 lru_gen_use_mm(struct mm_struct * mm)1087 static inline void lru_gen_use_mm(struct mm_struct *mm) 1088 { 1089 /* 1090 * When the bitmap is set, page reclaim knows this mm_struct has been 1091 * used since the last time it cleared the bitmap. So it might be worth 1092 * walking the page tables of this mm_struct to clear the accessed bit. 1093 */ 1094 WRITE_ONCE(mm->lru_gen.bitmap, -1); 1095 } 1096 1097 #else /* !CONFIG_LRU_GEN_WALKS_MMU */ 1098 lru_gen_add_mm(struct mm_struct * mm)1099 static inline void lru_gen_add_mm(struct mm_struct *mm) 1100 { 1101 } 1102 lru_gen_del_mm(struct mm_struct * mm)1103 static inline void lru_gen_del_mm(struct mm_struct *mm) 1104 { 1105 } 1106 lru_gen_migrate_mm(struct mm_struct * mm)1107 static inline void lru_gen_migrate_mm(struct mm_struct *mm) 1108 { 1109 } 1110 lru_gen_init_mm(struct mm_struct * mm)1111 static inline void lru_gen_init_mm(struct mm_struct *mm) 1112 { 1113 } 1114 lru_gen_use_mm(struct mm_struct * mm)1115 static inline void lru_gen_use_mm(struct mm_struct *mm) 1116 { 1117 } 1118 1119 #endif /* CONFIG_LRU_GEN_WALKS_MMU */ 1120 1121 struct vma_iterator { 1122 struct ma_state mas; 1123 }; 1124 1125 #define VMA_ITERATOR(name, __mm, __addr) \ 1126 struct vma_iterator name = { \ 1127 .mas = { \ 1128 .tree = &(__mm)->mm_mt, \ 1129 .index = __addr, \ 1130 .node = NULL, \ 1131 .status = ma_start, \ 1132 }, \ 1133 } 1134 vma_iter_init(struct vma_iterator * vmi,struct mm_struct * mm,unsigned long addr)1135 static inline void vma_iter_init(struct vma_iterator *vmi, 1136 struct mm_struct *mm, unsigned long addr) 1137 { 1138 mas_init(&vmi->mas, &mm->mm_mt, addr); 1139 } 1140 1141 #ifdef CONFIG_SCHED_MM_CID 1142 1143 enum mm_cid_state { 1144 MM_CID_UNSET = -1U, /* Unset state has lazy_put flag set. */ 1145 MM_CID_LAZY_PUT = (1U << 31), 1146 }; 1147 mm_cid_is_unset(int cid)1148 static inline bool mm_cid_is_unset(int cid) 1149 { 1150 return cid == MM_CID_UNSET; 1151 } 1152 mm_cid_is_lazy_put(int cid)1153 static inline bool mm_cid_is_lazy_put(int cid) 1154 { 1155 return !mm_cid_is_unset(cid) && (cid & MM_CID_LAZY_PUT); 1156 } 1157 mm_cid_is_valid(int cid)1158 static inline bool mm_cid_is_valid(int cid) 1159 { 1160 return !(cid & MM_CID_LAZY_PUT); 1161 } 1162 mm_cid_set_lazy_put(int cid)1163 static inline int mm_cid_set_lazy_put(int cid) 1164 { 1165 return cid | MM_CID_LAZY_PUT; 1166 } 1167 mm_cid_clear_lazy_put(int cid)1168 static inline int mm_cid_clear_lazy_put(int cid) 1169 { 1170 return cid & ~MM_CID_LAZY_PUT; 1171 } 1172 1173 /* Accessor for struct mm_struct's cidmask. */ mm_cidmask(struct mm_struct * mm)1174 static inline cpumask_t *mm_cidmask(struct mm_struct *mm) 1175 { 1176 unsigned long cid_bitmap = (unsigned long)mm; 1177 1178 cid_bitmap += offsetof(struct mm_struct, cpu_bitmap); 1179 /* Skip cpu_bitmap */ 1180 cid_bitmap += cpumask_size(); 1181 return (struct cpumask *)cid_bitmap; 1182 } 1183 mm_init_cid(struct mm_struct * mm)1184 static inline void mm_init_cid(struct mm_struct *mm) 1185 { 1186 int i; 1187 1188 for_each_possible_cpu(i) { 1189 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, i); 1190 1191 pcpu_cid->cid = MM_CID_UNSET; 1192 pcpu_cid->time = 0; 1193 } 1194 cpumask_clear(mm_cidmask(mm)); 1195 } 1196 mm_alloc_cid_noprof(struct mm_struct * mm)1197 static inline int mm_alloc_cid_noprof(struct mm_struct *mm) 1198 { 1199 mm->pcpu_cid = alloc_percpu_noprof(struct mm_cid); 1200 if (!mm->pcpu_cid) 1201 return -ENOMEM; 1202 mm_init_cid(mm); 1203 return 0; 1204 } 1205 #define mm_alloc_cid(...) alloc_hooks(mm_alloc_cid_noprof(__VA_ARGS__)) 1206 mm_destroy_cid(struct mm_struct * mm)1207 static inline void mm_destroy_cid(struct mm_struct *mm) 1208 { 1209 free_percpu(mm->pcpu_cid); 1210 mm->pcpu_cid = NULL; 1211 } 1212 mm_cid_size(void)1213 static inline unsigned int mm_cid_size(void) 1214 { 1215 return cpumask_size(); 1216 } 1217 #else /* CONFIG_SCHED_MM_CID */ mm_init_cid(struct mm_struct * mm)1218 static inline void mm_init_cid(struct mm_struct *mm) { } mm_alloc_cid(struct mm_struct * mm)1219 static inline int mm_alloc_cid(struct mm_struct *mm) { return 0; } mm_destroy_cid(struct mm_struct * mm)1220 static inline void mm_destroy_cid(struct mm_struct *mm) { } mm_cid_size(void)1221 static inline unsigned int mm_cid_size(void) 1222 { 1223 return 0; 1224 } 1225 #endif /* CONFIG_SCHED_MM_CID */ 1226 1227 struct mmu_gather; 1228 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm); 1229 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm); 1230 extern void tlb_finish_mmu(struct mmu_gather *tlb); 1231 1232 struct vm_fault; 1233 1234 /** 1235 * typedef vm_fault_t - Return type for page fault handlers. 1236 * 1237 * Page fault handlers return a bitmask of %VM_FAULT values. 1238 */ 1239 typedef __bitwise unsigned int vm_fault_t; 1240 1241 /** 1242 * enum vm_fault_reason - Page fault handlers return a bitmask of 1243 * these values to tell the core VM what happened when handling the 1244 * fault. Used to decide whether a process gets delivered SIGBUS or 1245 * just gets major/minor fault counters bumped up. 1246 * 1247 * @VM_FAULT_OOM: Out Of Memory 1248 * @VM_FAULT_SIGBUS: Bad access 1249 * @VM_FAULT_MAJOR: Page read from storage 1250 * @VM_FAULT_HWPOISON: Hit poisoned small page 1251 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded 1252 * in upper bits 1253 * @VM_FAULT_SIGSEGV: segmentation fault 1254 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page 1255 * @VM_FAULT_LOCKED: ->fault locked the returned page 1256 * @VM_FAULT_RETRY: ->fault blocked, must retry 1257 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small 1258 * @VM_FAULT_DONE_COW: ->fault has fully handled COW 1259 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs 1260 * fsync() to complete (for synchronous page faults 1261 * in DAX) 1262 * @VM_FAULT_COMPLETED: ->fault completed, meanwhile mmap lock released 1263 * @VM_FAULT_HINDEX_MASK: mask HINDEX value 1264 * 1265 */ 1266 enum vm_fault_reason { 1267 VM_FAULT_OOM = (__force vm_fault_t)0x000001, 1268 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002, 1269 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004, 1270 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010, 1271 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020, 1272 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040, 1273 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100, 1274 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200, 1275 VM_FAULT_RETRY = (__force vm_fault_t)0x000400, 1276 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800, 1277 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000, 1278 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000, 1279 VM_FAULT_COMPLETED = (__force vm_fault_t)0x004000, 1280 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000, 1281 }; 1282 1283 /* Encode hstate index for a hwpoisoned large page */ 1284 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16)) 1285 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf) 1286 1287 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \ 1288 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \ 1289 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK) 1290 1291 #define VM_FAULT_RESULT_TRACE \ 1292 { VM_FAULT_OOM, "OOM" }, \ 1293 { VM_FAULT_SIGBUS, "SIGBUS" }, \ 1294 { VM_FAULT_MAJOR, "MAJOR" }, \ 1295 { VM_FAULT_HWPOISON, "HWPOISON" }, \ 1296 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \ 1297 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \ 1298 { VM_FAULT_NOPAGE, "NOPAGE" }, \ 1299 { VM_FAULT_LOCKED, "LOCKED" }, \ 1300 { VM_FAULT_RETRY, "RETRY" }, \ 1301 { VM_FAULT_FALLBACK, "FALLBACK" }, \ 1302 { VM_FAULT_DONE_COW, "DONE_COW" }, \ 1303 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }, \ 1304 { VM_FAULT_COMPLETED, "COMPLETED" } 1305 1306 struct vm_special_mapping { 1307 const char *name; /* The name, e.g. "[vdso]". */ 1308 1309 /* 1310 * If .fault is not provided, this points to a 1311 * NULL-terminated array of pages that back the special mapping. 1312 * 1313 * This must not be NULL unless .fault is provided. 1314 */ 1315 struct page **pages; 1316 1317 /* 1318 * If non-NULL, then this is called to resolve page faults 1319 * on the special mapping. If used, .pages is not checked. 1320 */ 1321 vm_fault_t (*fault)(const struct vm_special_mapping *sm, 1322 struct vm_area_struct *vma, 1323 struct vm_fault *vmf); 1324 1325 int (*mremap)(const struct vm_special_mapping *sm, 1326 struct vm_area_struct *new_vma); 1327 1328 void (*close)(const struct vm_special_mapping *sm, 1329 struct vm_area_struct *vma); 1330 }; 1331 1332 enum tlb_flush_reason { 1333 TLB_FLUSH_ON_TASK_SWITCH, 1334 TLB_REMOTE_SHOOTDOWN, 1335 TLB_LOCAL_SHOOTDOWN, 1336 TLB_LOCAL_MM_SHOOTDOWN, 1337 TLB_REMOTE_SEND_IPI, 1338 NR_TLB_FLUSH_REASONS, 1339 }; 1340 1341 /** 1342 * enum fault_flag - Fault flag definitions. 1343 * @FAULT_FLAG_WRITE: Fault was a write fault. 1344 * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE. 1345 * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked. 1346 * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying. 1347 * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region. 1348 * @FAULT_FLAG_TRIED: The fault has been tried once. 1349 * @FAULT_FLAG_USER: The fault originated in userspace. 1350 * @FAULT_FLAG_REMOTE: The fault is not for current task/mm. 1351 * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch. 1352 * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals. 1353 * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a 1354 * COW mapping, making sure that an exclusive anon page is 1355 * mapped after the fault. 1356 * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached. 1357 * We should only access orig_pte if this flag set. 1358 * @FAULT_FLAG_VMA_LOCK: The fault is handled under VMA lock. 1359 * 1360 * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify 1361 * whether we would allow page faults to retry by specifying these two 1362 * fault flags correctly. Currently there can be three legal combinations: 1363 * 1364 * (a) ALLOW_RETRY and !TRIED: this means the page fault allows retry, and 1365 * this is the first try 1366 * 1367 * (b) ALLOW_RETRY and TRIED: this means the page fault allows retry, and 1368 * we've already tried at least once 1369 * 1370 * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry 1371 * 1372 * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never 1373 * be used. Note that page faults can be allowed to retry for multiple times, 1374 * in which case we'll have an initial fault with flags (a) then later on 1375 * continuous faults with flags (b). We should always try to detect pending 1376 * signals before a retry to make sure the continuous page faults can still be 1377 * interrupted if necessary. 1378 * 1379 * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal. 1380 * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when 1381 * applied to mappings that are not COW mappings. 1382 */ 1383 enum fault_flag { 1384 FAULT_FLAG_WRITE = 1 << 0, 1385 FAULT_FLAG_MKWRITE = 1 << 1, 1386 FAULT_FLAG_ALLOW_RETRY = 1 << 2, 1387 FAULT_FLAG_RETRY_NOWAIT = 1 << 3, 1388 FAULT_FLAG_KILLABLE = 1 << 4, 1389 FAULT_FLAG_TRIED = 1 << 5, 1390 FAULT_FLAG_USER = 1 << 6, 1391 FAULT_FLAG_REMOTE = 1 << 7, 1392 FAULT_FLAG_INSTRUCTION = 1 << 8, 1393 FAULT_FLAG_INTERRUPTIBLE = 1 << 9, 1394 FAULT_FLAG_UNSHARE = 1 << 10, 1395 FAULT_FLAG_ORIG_PTE_VALID = 1 << 11, 1396 FAULT_FLAG_VMA_LOCK = 1 << 12, 1397 }; 1398 1399 typedef unsigned int __bitwise zap_flags_t; 1400 1401 /* Flags for clear_young_dirty_ptes(). */ 1402 typedef int __bitwise cydp_t; 1403 1404 /* Clear the access bit */ 1405 #define CYDP_CLEAR_YOUNG ((__force cydp_t)BIT(0)) 1406 1407 /* Clear the dirty bit */ 1408 #define CYDP_CLEAR_DIRTY ((__force cydp_t)BIT(1)) 1409 1410 /* 1411 * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each 1412 * other. Here is what they mean, and how to use them: 1413 * 1414 * 1415 * FIXME: For pages which are part of a filesystem, mappings are subject to the 1416 * lifetime enforced by the filesystem and we need guarantees that longterm 1417 * users like RDMA and V4L2 only establish mappings which coordinate usage with 1418 * the filesystem. Ideas for this coordination include revoking the longterm 1419 * pin, delaying writeback, bounce buffer page writeback, etc. As FS DAX was 1420 * added after the problem with filesystems was found FS DAX VMAs are 1421 * specifically failed. Filesystem pages are still subject to bugs and use of 1422 * FOLL_LONGTERM should be avoided on those pages. 1423 * 1424 * In the CMA case: long term pins in a CMA region would unnecessarily fragment 1425 * that region. And so, CMA attempts to migrate the page before pinning, when 1426 * FOLL_LONGTERM is specified. 1427 * 1428 * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount, 1429 * but an additional pin counting system) will be invoked. This is intended for 1430 * anything that gets a page reference and then touches page data (for example, 1431 * Direct IO). This lets the filesystem know that some non-file-system entity is 1432 * potentially changing the pages' data. In contrast to FOLL_GET (whose pages 1433 * are released via put_page()), FOLL_PIN pages must be released, ultimately, by 1434 * a call to unpin_user_page(). 1435 * 1436 * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different 1437 * and separate refcounting mechanisms, however, and that means that each has 1438 * its own acquire and release mechanisms: 1439 * 1440 * FOLL_GET: get_user_pages*() to acquire, and put_page() to release. 1441 * 1442 * FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release. 1443 * 1444 * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call. 1445 * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based 1446 * calls applied to them, and that's perfectly OK. This is a constraint on the 1447 * callers, not on the pages.) 1448 * 1449 * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never 1450 * directly by the caller. That's in order to help avoid mismatches when 1451 * releasing pages: get_user_pages*() pages must be released via put_page(), 1452 * while pin_user_pages*() pages must be released via unpin_user_page(). 1453 * 1454 * Please see Documentation/core-api/pin_user_pages.rst for more information. 1455 */ 1456 1457 enum { 1458 /* check pte is writable */ 1459 FOLL_WRITE = 1 << 0, 1460 /* do get_page on page */ 1461 FOLL_GET = 1 << 1, 1462 /* give error on hole if it would be zero */ 1463 FOLL_DUMP = 1 << 2, 1464 /* get_user_pages read/write w/o permission */ 1465 FOLL_FORCE = 1 << 3, 1466 /* 1467 * if a disk transfer is needed, start the IO and return without waiting 1468 * upon it 1469 */ 1470 FOLL_NOWAIT = 1 << 4, 1471 /* do not fault in pages */ 1472 FOLL_NOFAULT = 1 << 5, 1473 /* check page is hwpoisoned */ 1474 FOLL_HWPOISON = 1 << 6, 1475 /* don't do file mappings */ 1476 FOLL_ANON = 1 << 7, 1477 /* 1478 * FOLL_LONGTERM indicates that the page will be held for an indefinite 1479 * time period _often_ under userspace control. This is in contrast to 1480 * iov_iter_get_pages(), whose usages are transient. 1481 */ 1482 FOLL_LONGTERM = 1 << 8, 1483 /* split huge pmd before returning */ 1484 FOLL_SPLIT_PMD = 1 << 9, 1485 /* allow returning PCI P2PDMA pages */ 1486 FOLL_PCI_P2PDMA = 1 << 10, 1487 /* allow interrupts from generic signals */ 1488 FOLL_INTERRUPTIBLE = 1 << 11, 1489 /* 1490 * Always honor (trigger) NUMA hinting faults. 1491 * 1492 * FOLL_WRITE implicitly honors NUMA hinting faults because a 1493 * PROT_NONE-mapped page is not writable (exceptions with FOLL_FORCE 1494 * apply). get_user_pages_fast_only() always implicitly honors NUMA 1495 * hinting faults. 1496 */ 1497 FOLL_HONOR_NUMA_FAULT = 1 << 12, 1498 1499 /* See also internal only FOLL flags in mm/internal.h */ 1500 }; 1501 1502 #endif /* _LINUX_MM_TYPES_H */ 1503