mm_types.h

   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#include <linux/types.h>
  23#include <linux/rseq_types.h>
  24#include <linux/bitmap.h>
  25
  26#include <asm/mmu.h>
  27
  28#ifndef AT_VECTOR_SIZE_ARCH
  29#define AT_VECTOR_SIZE_ARCH 0
  30#endif
  31#define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
  32
  33
  34struct address_space;
  35struct futex_private_hash;
  36struct mem_cgroup;
  37
  38typedef struct {
  39	unsigned long f;
  40} memdesc_flags_t;
  41
  42/*
  43 * Each physical page in the system has a struct page associated with
  44 * it to keep track of whatever it is we are using the page for at the
  45 * moment. Note that we have no way to track which tasks are using
  46 * a page, though if it is a pagecache page, rmap structures can tell us
  47 * who is mapping it.
  48 *
  49 * If you allocate the page using alloc_pages(), you can use some of the
  50 * space in struct page for your own purposes.  The five words in the main
  51 * union are available, except for bit 0 of the first word which must be
  52 * kept clear.  Many users use this word to store a pointer to an object
  53 * which is guaranteed to be aligned.  If you use the same storage as
  54 * page->mapping, you must restore it to NULL before freeing the page.
  55 *
  56 * The mapcount field must not be used for own purposes.
  57 *
  58 * If you want to use the refcount field, it must be used in such a way
  59 * that other CPUs temporarily incrementing and then decrementing the
  60 * refcount does not cause problems.  On receiving the page from
  61 * alloc_pages(), the refcount will be positive.
  62 *
  63 * If you allocate pages of order > 0, you can use some of the fields
  64 * in each subpage, but you may need to restore some of their values
  65 * afterwards.
  66 *
  67 * SLUB uses cmpxchg_double() to atomically update its freelist and counters.
  68 * That requires that freelist & counters in struct slab be adjacent and
  69 * double-word aligned. Because struct slab currently just reinterprets the
  70 * bits of struct page, we align all struct pages to double-word boundaries,
  71 * and ensure that 'freelist' is aligned within struct slab.
  72 */
  73#ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
  74#define _struct_page_alignment	__aligned(2 * sizeof(unsigned long))
  75#else
  76#define _struct_page_alignment	__aligned(sizeof(unsigned long))
  77#endif
  78
  79struct page {
  80	memdesc_flags_t flags;		/* Atomic flags, some possibly
  81					 * updated asynchronously */
  82	/*
  83	 * Five words (20/40 bytes) are available in this union.
  84	 * WARNING: bit 0 of the first word is used for PageTail(). That
  85	 * means the other users of this union MUST NOT use the bit to
  86	 * avoid collision and false-positive PageTail().
  87	 */
  88	union {
  89		struct {	/* Page cache and anonymous pages */
  90			/**
  91			 * @lru: Pageout list, eg. active_list protected by
  92			 * lruvec->lru_lock.  Sometimes used as a generic list
  93			 * by the page owner.
  94			 */
  95			union {
  96				struct list_head lru;
  97
  98				/* Or, free page */
  99				struct list_head buddy_list;
 100				struct list_head pcp_list;
 101				struct llist_node pcp_llist;
 102			};
 103			struct address_space *mapping;
 104			union {
 105				pgoff_t __folio_index;		/* Our offset within mapping. */
 106				unsigned long share;	/* share count for fsdax */
 107			};
 108			/**
 109			 * @private: Mapping-private opaque data.
 110			 * Usually used for buffer_heads if PagePrivate.
 111			 * Used for swp_entry_t if swapcache flag set.
 112			 * Indicates order in the buddy system if PageBuddy
 113			 * or on pcp_llist.
 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			/*
 133			 * The first word is used for compound_head or folio
 134			 * pgmap
 135			 */
 136			void *_unused_pgmap_compound_head;
 137			void *zone_device_data;
 138			/*
 139			 * ZONE_DEVICE private pages are counted as being
 140			 * mapped so the next 3 words hold the mapping, index,
 141			 * and private fields from the source anonymous or
 142			 * page cache page while the page is migrated to device
 143			 * private memory.
 144			 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
 145			 * use the mapping, index, and private fields when
 146			 * pmem backed DAX files are mapped.
 147			 */
 148		};
 149
 150		/** @rcu_head: You can use this to free a page by RCU. */
 151		struct rcu_head rcu_head;
 152	};
 153
 154	union {		/* This union is 4 bytes in size. */
 155		/*
 156		 * For head pages of typed folios, the value stored here
 157		 * allows for determining what this page is used for. The
 158		 * tail pages of typed folios will not store a type
 159		 * (page_type == _mapcount == -1).
 160		 *
 161		 * See page-flags.h for a list of page types which are currently
 162		 * stored here.
 163		 *
 164		 * Owners of typed folios may reuse the lower 16 bit of the
 165		 * head page page_type field after setting the page type,
 166		 * but must reset these 16 bit to -1 before clearing the
 167		 * page type.
 168		 */
 169		unsigned int page_type;
 170
 171		/*
 172		 * For pages that are part of non-typed folios for which mappings
 173		 * are tracked via the RMAP, encodes the number of times this page
 174		 * is directly referenced by a page table.
 175		 *
 176		 * Note that the mapcount is always initialized to -1, so that
 177		 * transitions both from it and to it can be tracked, using
 178		 * atomic_inc_and_test() and atomic_add_negative(-1).
 179		 */
 180		atomic_t _mapcount;
 181	};
 182
 183	/* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
 184	atomic_t _refcount;
 185
 186#ifdef CONFIG_MEMCG
 187	unsigned long memcg_data;
 188#elif defined(CONFIG_SLAB_OBJ_EXT)
 189	unsigned long _unused_slab_obj_exts;
 190#endif
 191
 192	/*
 193	 * On machines where all RAM is mapped into kernel address space,
 194	 * we can simply calculate the virtual address. On machines with
 195	 * highmem some memory is mapped into kernel virtual memory
 196	 * dynamically, so we need a place to store that address.
 197	 * Note that this field could be 16 bits on x86 ... ;)
 198	 *
 199	 * Architectures with slow multiplication can define
 200	 * WANT_PAGE_VIRTUAL in asm/page.h
 201	 */
 202#if defined(WANT_PAGE_VIRTUAL)
 203	void *virtual;			/* Kernel virtual address (NULL if
 204					   not kmapped, ie. highmem) */
 205#endif /* WANT_PAGE_VIRTUAL */
 206
 207#ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 208	int _last_cpupid;
 209#endif
 210
 211#ifdef CONFIG_KMSAN
 212	/*
 213	 * KMSAN metadata for this page:
 214	 *  - shadow page: every bit indicates whether the corresponding
 215	 *    bit of the original page is initialized (0) or not (1);
 216	 *  - origin page: every 4 bytes contain an id of the stack trace
 217	 *    where the uninitialized value was created.
 218	 */
 219	struct page *kmsan_shadow;
 220	struct page *kmsan_origin;
 221#endif
 222} _struct_page_alignment;
 223
 224/*
 225 * struct encoded_page - a nonexistent type marking this pointer
 226 *
 227 * An 'encoded_page' pointer is a pointer to a regular 'struct page', but
 228 * with the low bits of the pointer indicating extra context-dependent
 229 * information. Only used in mmu_gather handling, and this acts as a type
 230 * system check on that use.
 231 *
 232 * We only really have two guaranteed bits in general, although you could
 233 * play with 'struct page' alignment (see CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
 234 * for more.
 235 *
 236 * Use the supplied helper functions to endcode/decode the pointer and bits.
 237 */
 238struct encoded_page;
 239
 240#define ENCODED_PAGE_BITS			3ul
 241
 242/* Perform rmap removal after we have flushed the TLB. */
 243#define ENCODED_PAGE_BIT_DELAY_RMAP		1ul
 244
 245/*
 246 * The next item in an encoded_page array is the "nr_pages" argument, specifying
 247 * the number of consecutive pages starting from this page, that all belong to
 248 * the same folio. For example, "nr_pages" corresponds to the number of folio
 249 * references that must be dropped. If this bit is not set, "nr_pages" is
 250 * implicitly 1.
 251 */
 252#define ENCODED_PAGE_BIT_NR_PAGES_NEXT		2ul
 253
 254static __always_inline struct encoded_page *encode_page(struct page *page, unsigned long flags)
 255{
 256	BUILD_BUG_ON(flags > ENCODED_PAGE_BITS);
 257	return (struct encoded_page *)(flags | (unsigned long)page);
 258}
 259
 260static inline unsigned long encoded_page_flags(struct encoded_page *page)
 261{
 262	return ENCODED_PAGE_BITS & (unsigned long)page;
 263}
 264
 265static inline struct page *encoded_page_ptr(struct encoded_page *page)
 266{
 267	return (struct page *)(~ENCODED_PAGE_BITS & (unsigned long)page);
 268}
 269
 270static __always_inline struct encoded_page *encode_nr_pages(unsigned long nr)
 271{
 272	VM_WARN_ON_ONCE((nr << 2) >> 2 != nr);
 273	return (struct encoded_page *)(nr << 2);
 274}
 275
 276static __always_inline unsigned long encoded_nr_pages(struct encoded_page *page)
 277{
 278	return ((unsigned long)page) >> 2;
 279}
 280
 281/*
 282 * A swap entry has to fit into a "unsigned long", as the entry is hidden
 283 * in the "index" field of the swapper address space.
 284 */
 285typedef struct {
 286	unsigned long val;
 287} swp_entry_t;
 288
 289/**
 290 * typedef softleaf_t - Describes a page table software leaf entry, abstracted
 291 * from its architecture-specific encoding.
 292 *
 293 * Page table leaf entries are those which do not reference any descendent page
 294 * tables but rather either reference a data page, are an empty (or 'none'
 295 * entry), or contain a non-present entry.
 296 *
 297 * If referencing another page table or a data page then the page table entry is
 298 * pertinent to hardware - that is it tells the hardware how to decode the page
 299 * table entry.
 300 *
 301 * Otherwise it is a software-defined leaf page table entry, which this type
 302 * describes. See leafops.h and specifically @softleaf_type for a list of all
 303 * possible kinds of software leaf entry.
 304 *
 305 * A softleaf_t entry is abstracted from the hardware page table entry, so is
 306 * not architecture-specific.
 307 *
 308 * NOTE: While we transition from the confusing swp_entry_t type used for this
 309 *       purpose, we simply alias this type. This will be removed once the
 310 *       transition is complete.
 311 */
 312typedef swp_entry_t softleaf_t;
 313
 314#if defined(CONFIG_MEMCG) || defined(CONFIG_SLAB_OBJ_EXT)
 315/* We have some extra room after the refcount in tail pages. */
 316#define NR_PAGES_IN_LARGE_FOLIO
 317#endif
 318
 319/*
 320 * On 32bit, we can cut the required metadata in half, because:
 321 * (a) PID_MAX_LIMIT implicitly limits the number of MMs we could ever have,
 322 *     so we can limit MM IDs to 15 bit (32767).
 323 * (b) We don't expect folios where even a single complete PTE mapping by
 324 *     one MM would exceed 15 bits (order-15).
 325 */
 326#ifdef CONFIG_64BIT
 327typedef int mm_id_mapcount_t;
 328#define MM_ID_MAPCOUNT_MAX		INT_MAX
 329typedef unsigned int mm_id_t;
 330#else /* !CONFIG_64BIT */
 331typedef short mm_id_mapcount_t;
 332#define MM_ID_MAPCOUNT_MAX		SHRT_MAX
 333typedef unsigned short mm_id_t;
 334#endif /* CONFIG_64BIT */
 335
 336/* We implicitly use the dummy ID for init-mm etc. where we never rmap pages. */
 337#define MM_ID_DUMMY			0
 338#define MM_ID_MIN			(MM_ID_DUMMY + 1)
 339
 340/*
 341 * We leave the highest bit of each MM id unused, so we can store a flag
 342 * in the highest bit of each folio->_mm_id[].
 343 */
 344#define MM_ID_BITS			((sizeof(mm_id_t) * BITS_PER_BYTE) - 1)
 345#define MM_ID_MASK			((1U << MM_ID_BITS) - 1)
 346#define MM_ID_MAX			MM_ID_MASK
 347
 348/*
 349 * In order to use bit_spin_lock(), which requires an unsigned long, we
 350 * operate on folio->_mm_ids when working on flags.
 351 */
 352#define FOLIO_MM_IDS_LOCK_BITNUM	MM_ID_BITS
 353#define FOLIO_MM_IDS_LOCK_BIT		BIT(FOLIO_MM_IDS_LOCK_BITNUM)
 354#define FOLIO_MM_IDS_SHARED_BITNUM	(2 * MM_ID_BITS + 1)
 355#define FOLIO_MM_IDS_SHARED_BIT		BIT(FOLIO_MM_IDS_SHARED_BITNUM)
 356
 357/**
 358 * struct folio - Represents a contiguous set of bytes.
 359 * @flags: Identical to the page flags.
 360 * @lru: Least Recently Used list; tracks how recently this folio was used.
 361 * @mlock_count: Number of times this folio has been pinned by mlock().
 362 * @mapping: The file this page belongs to, or refers to the anon_vma for
 363 *    anonymous memory.
 364 * @index: Offset within the file, in units of pages.  For anonymous memory,
 365 *    this is the index from the beginning of the mmap.
 366 * @share: number of DAX mappings that reference this folio. See
 367 *    dax_associate_entry.
 368 * @private: Filesystem per-folio data (see folio_attach_private()).
 369 * @swap: Used for swp_entry_t if folio_test_swapcache().
 370 * @_mapcount: Do not access this member directly.  Use folio_mapcount() to
 371 *    find out how many times this folio is mapped by userspace.
 372 * @_refcount: Do not access this member directly.  Use folio_ref_count()
 373 *    to find how many references there are to this folio.
 374 * @memcg_data: Memory Control Group data.
 375 * @pgmap: Metadata for ZONE_DEVICE mappings
 376 * @virtual: Virtual address in the kernel direct map.
 377 * @_last_cpupid: IDs of last CPU and last process that accessed the folio.
 378 * @_entire_mapcount: Do not use directly, call folio_entire_mapcount().
 379 * @_large_mapcount: Do not use directly, call folio_mapcount().
 380 * @_nr_pages_mapped: Do not use outside of rmap and debug code.
 381 * @_pincount: Do not use directly, call folio_maybe_dma_pinned().
 382 * @_nr_pages: Do not use directly, call folio_nr_pages().
 383 * @_mm_id: Do not use outside of rmap code.
 384 * @_mm_ids: Do not use outside of rmap code.
 385 * @_mm_id_mapcount: Do not use outside of rmap code.
 386 * @_hugetlb_subpool: Do not use directly, use accessor in hugetlb.h.
 387 * @_hugetlb_cgroup: Do not use directly, use accessor in hugetlb_cgroup.h.
 388 * @_hugetlb_cgroup_rsvd: Do not use directly, use accessor in hugetlb_cgroup.h.
 389 * @_hugetlb_hwpoison: Do not use directly, call raw_hwp_list_head().
 390 * @_deferred_list: Folios to be split under memory pressure.
 391 * @_unused_slab_obj_exts: Placeholder to match obj_exts in struct slab.
 392 *
 393 * A folio is a physically, virtually and logically contiguous set
 394 * of bytes.  It is a power-of-two in size, and it is aligned to that
 395 * same power-of-two.  It is at least as large as %PAGE_SIZE.  If it is
 396 * in the page cache, it is at a file offset which is a multiple of that
 397 * power-of-two.  It may be mapped into userspace at an address which is
 398 * at an arbitrary page offset, but its kernel virtual address is aligned
 399 * to its size.
 400 */
 401struct folio {
 402	/* private: don't document the anon union */
 403	union {
 404		struct {
 405	/* public: */
 406			memdesc_flags_t flags;
 407			union {
 408				struct list_head lru;
 409	/* private: avoid cluttering the output */
 410				/* For the Unevictable "LRU list" slot */
 411				struct {
 412					/* Avoid compound_head */
 413					void *__filler;
 414	/* public: */
 415					unsigned int mlock_count;
 416	/* private: */
 417				};
 418	/* public: */
 419				struct dev_pagemap *pgmap;
 420			};
 421			struct address_space *mapping;
 422			union {
 423				pgoff_t index;
 424				unsigned long share;
 425			};
 426			union {
 427				void *private;
 428				swp_entry_t swap;
 429			};
 430			atomic_t _mapcount;
 431			atomic_t _refcount;
 432#ifdef CONFIG_MEMCG
 433			unsigned long memcg_data;
 434#elif defined(CONFIG_SLAB_OBJ_EXT)
 435			unsigned long _unused_slab_obj_exts;
 436#endif
 437#if defined(WANT_PAGE_VIRTUAL)
 438			void *virtual;
 439#endif
 440#ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 441			int _last_cpupid;
 442#endif
 443	/* private: the union with struct page is transitional */
 444		};
 445		struct page page;
 446	};
 447	union {
 448		struct {
 449			unsigned long _flags_1;
 450			unsigned long _head_1;
 451			union {
 452				struct {
 453	/* public: */
 454					atomic_t _large_mapcount;
 455					atomic_t _nr_pages_mapped;
 456#ifdef CONFIG_64BIT
 457					atomic_t _entire_mapcount;
 458					atomic_t _pincount;
 459#endif /* CONFIG_64BIT */
 460					mm_id_mapcount_t _mm_id_mapcount[2];
 461					union {
 462						mm_id_t _mm_id[2];
 463						unsigned long _mm_ids;
 464					};
 465	/* private: the union with struct page is transitional */
 466				};
 467				unsigned long _usable_1[4];
 468			};
 469			atomic_t _mapcount_1;
 470			atomic_t _refcount_1;
 471	/* public: */
 472#ifdef NR_PAGES_IN_LARGE_FOLIO
 473			unsigned int _nr_pages;
 474#endif /* NR_PAGES_IN_LARGE_FOLIO */
 475	/* private: the union with struct page is transitional */
 476		};
 477		struct page __page_1;
 478	};
 479	union {
 480		struct {
 481			unsigned long _flags_2;
 482			unsigned long _head_2;
 483	/* public: */
 484			struct list_head _deferred_list;
 485#ifndef CONFIG_64BIT
 486			atomic_t _entire_mapcount;
 487			atomic_t _pincount;
 488#endif /* !CONFIG_64BIT */
 489	/* private: the union with struct page is transitional */
 490		};
 491		struct page __page_2;
 492	};
 493	union {
 494		struct {
 495			unsigned long _flags_3;
 496			unsigned long _head_3;
 497	/* public: */
 498			void *_hugetlb_subpool;
 499			void *_hugetlb_cgroup;
 500			void *_hugetlb_cgroup_rsvd;
 501			void *_hugetlb_hwpoison;
 502	/* private: the union with struct page is transitional */
 503		};
 504		struct page __page_3;
 505	};
 506};
 507
 508#define FOLIO_MATCH(pg, fl)						\
 509	static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
 510FOLIO_MATCH(flags, flags);
 511FOLIO_MATCH(lru, lru);
 512FOLIO_MATCH(mapping, mapping);
 513FOLIO_MATCH(compound_head, lru);
 514FOLIO_MATCH(__folio_index, index);
 515FOLIO_MATCH(private, private);
 516FOLIO_MATCH(_mapcount, _mapcount);
 517FOLIO_MATCH(_refcount, _refcount);
 518#ifdef CONFIG_MEMCG
 519FOLIO_MATCH(memcg_data, memcg_data);
 520#endif
 521#if defined(WANT_PAGE_VIRTUAL)
 522FOLIO_MATCH(virtual, virtual);
 523#endif
 524#ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 525FOLIO_MATCH(_last_cpupid, _last_cpupid);
 526#endif
 527#undef FOLIO_MATCH
 528#define FOLIO_MATCH(pg, fl)						\
 529	static_assert(offsetof(struct folio, fl) ==			\
 530			offsetof(struct page, pg) + sizeof(struct page))
 531FOLIO_MATCH(flags, _flags_1);
 532FOLIO_MATCH(compound_head, _head_1);
 533FOLIO_MATCH(_mapcount, _mapcount_1);
 534FOLIO_MATCH(_refcount, _refcount_1);
 535#undef FOLIO_MATCH
 536#define FOLIO_MATCH(pg, fl)						\
 537	static_assert(offsetof(struct folio, fl) ==			\
 538			offsetof(struct page, pg) + 2 * sizeof(struct page))
 539FOLIO_MATCH(flags, _flags_2);
 540FOLIO_MATCH(compound_head, _head_2);
 541#undef FOLIO_MATCH
 542#define FOLIO_MATCH(pg, fl)						\
 543	static_assert(offsetof(struct folio, fl) ==			\
 544			offsetof(struct page, pg) + 3 * sizeof(struct page))
 545FOLIO_MATCH(flags, _flags_3);
 546FOLIO_MATCH(compound_head, _head_3);
 547#undef FOLIO_MATCH
 548
 549/**
 550 * struct ptdesc -    Memory descriptor for page tables.
 551 * @pt_flags: enum pt_flags plus zone/node/section.
 552 * @pt_rcu_head:      For freeing page table pages.
 553 * @pt_list:          List of used page tables. Used for s390 gmap shadow pages
 554 *                    (which are not linked into the user page tables) and x86
 555 *                    pgds.
 556 * @_pt_pad_1:        Padding that aliases with page's compound head.
 557 * @pmd_huge_pte:     Protected by ptdesc->ptl, used for THPs.
 558 * @__page_mapping:   Aliases with page->mapping. Unused for page tables.
 559 * @pt_index:         Used for s390 gmap.
 560 * @pt_mm:            Used for x86 pgds.
 561 * @pt_frag_refcount: For fragmented page table tracking. Powerpc only.
 562 * @pt_share_count:   Used for HugeTLB PMD page table share count.
 563 * @_pt_pad_2:        Padding to ensure proper alignment.
 564 * @ptl:              Lock for the page table.
 565 * @__page_type:      Same as page->page_type. Unused for page tables.
 566 * @__page_refcount:  Same as page refcount.
 567 * @pt_memcg_data:    Memcg data. Tracked for page tables here.
 568 *
 569 * This struct overlays struct page for now. Do not modify without a good
 570 * understanding of the issues.
 571 */
 572struct ptdesc {
 573	memdesc_flags_t pt_flags;
 574
 575	union {
 576		struct rcu_head pt_rcu_head;
 577		struct list_head pt_list;
 578		struct {
 579			unsigned long _pt_pad_1;
 580			pgtable_t pmd_huge_pte;
 581		};
 582	};
 583	unsigned long __page_mapping;
 584
 585	union {
 586		pgoff_t pt_index;
 587		struct mm_struct *pt_mm;
 588		atomic_t pt_frag_refcount;
 589#ifdef CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING
 590		atomic_t pt_share_count;
 591#endif
 592	};
 593
 594	union {
 595		unsigned long _pt_pad_2;
 596#if ALLOC_SPLIT_PTLOCKS
 597		spinlock_t *ptl;
 598#else
 599		spinlock_t ptl;
 600#endif
 601	};
 602	unsigned int __page_type;
 603	atomic_t __page_refcount;
 604#ifdef CONFIG_MEMCG
 605	unsigned long pt_memcg_data;
 606#endif
 607};
 608
 609#define TABLE_MATCH(pg, pt)						\
 610	static_assert(offsetof(struct page, pg) == offsetof(struct ptdesc, pt))
 611TABLE_MATCH(flags, pt_flags);
 612TABLE_MATCH(compound_head, pt_list);
 613TABLE_MATCH(compound_head, _pt_pad_1);
 614TABLE_MATCH(mapping, __page_mapping);
 615TABLE_MATCH(__folio_index, pt_index);
 616TABLE_MATCH(rcu_head, pt_rcu_head);
 617TABLE_MATCH(page_type, __page_type);
 618TABLE_MATCH(_refcount, __page_refcount);
 619#ifdef CONFIG_MEMCG
 620TABLE_MATCH(memcg_data, pt_memcg_data);
 621#endif
 622#undef TABLE_MATCH
 623static_assert(sizeof(struct ptdesc) <= sizeof(struct page));
 624
 625#define ptdesc_page(pt)			(_Generic((pt),			\
 626	const struct ptdesc *:		(const struct page *)(pt),	\
 627	struct ptdesc *:		(struct page *)(pt)))
 628
 629#define ptdesc_folio(pt)		(_Generic((pt),			\
 630	const struct ptdesc *:		(const struct folio *)(pt),	\
 631	struct ptdesc *:		(struct folio *)(pt)))
 632
 633#define page_ptdesc(p)			(_Generic((p),			\
 634	const struct page *:		(const struct ptdesc *)(p),	\
 635	struct page *:			(struct ptdesc *)(p)))
 636
 637#ifdef CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING
 638static inline void ptdesc_pmd_pts_init(struct ptdesc *ptdesc)
 639{
 640	atomic_set(&ptdesc->pt_share_count, 0);
 641}
 642
 643static inline void ptdesc_pmd_pts_inc(struct ptdesc *ptdesc)
 644{
 645	atomic_inc(&ptdesc->pt_share_count);
 646}
 647
 648static inline void ptdesc_pmd_pts_dec(struct ptdesc *ptdesc)
 649{
 650	atomic_dec(&ptdesc->pt_share_count);
 651}
 652
 653static inline int ptdesc_pmd_pts_count(const struct ptdesc *ptdesc)
 654{
 655	return atomic_read(&ptdesc->pt_share_count);
 656}
 657
 658static inline bool ptdesc_pmd_is_shared(struct ptdesc *ptdesc)
 659{
 660	return !!ptdesc_pmd_pts_count(ptdesc);
 661}
 662#else
 663static inline void ptdesc_pmd_pts_init(struct ptdesc *ptdesc)
 664{
 665}
 666#endif
 667
 668/*
 669 * Used for sizing the vmemmap region on some architectures
 670 */
 671#define STRUCT_PAGE_MAX_SHIFT	(order_base_2(sizeof(struct page)))
 672
 673/*
 674 * page_private can be used on tail pages.  However, PagePrivate is only
 675 * checked by the VM on the head page.  So page_private on the tail pages
 676 * should be used for data that's ancillary to the head page (eg attaching
 677 * buffer heads to tail pages after attaching buffer heads to the head page)
 678 */
 679#define page_private(page)		((page)->private)
 680
 681static inline void set_page_private(struct page *page, unsigned long private)
 682{
 683	page->private = private;
 684}
 685
 686static inline void *folio_get_private(const struct folio *folio)
 687{
 688	return folio->private;
 689}
 690
 691typedef unsigned long vm_flags_t;
 692
 693/*
 694 * freeptr_t represents a SLUB freelist pointer, which might be encoded
 695 * and not dereferenceable if CONFIG_SLAB_FREELIST_HARDENED is enabled.
 696 */
 697typedef struct { unsigned long v; } freeptr_t;
 698
 699/*
 700 * A region containing a mapping of a non-memory backed file under NOMMU
 701 * conditions.  These are held in a global tree and are pinned by the VMAs that
 702 * map parts of them.
 703 */
 704struct vm_region {
 705	struct rb_node	vm_rb;		/* link in global region tree */
 706	vm_flags_t	vm_flags;	/* VMA vm_flags */
 707	unsigned long	vm_start;	/* start address of region */
 708	unsigned long	vm_end;		/* region initialised to here */
 709	unsigned long	vm_top;		/* region allocated to here */
 710	unsigned long	vm_pgoff;	/* the offset in vm_file corresponding to vm_start */
 711	struct file	*vm_file;	/* the backing file or NULL */
 712
 713	int		vm_usage;	/* region usage count (access under nommu_region_sem) */
 714	bool		vm_icache_flushed : 1; /* true if the icache has been flushed for
 715						* this region */
 716};
 717
 718#ifdef CONFIG_USERFAULTFD
 719#define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
 720struct vm_userfaultfd_ctx {
 721	struct userfaultfd_ctx *ctx;
 722};
 723#else /* CONFIG_USERFAULTFD */
 724#define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
 725struct vm_userfaultfd_ctx {};
 726#endif /* CONFIG_USERFAULTFD */
 727
 728struct anon_vma_name {
 729	struct kref kref;
 730	/* The name needs to be at the end because it is dynamically sized. */
 731	char name[];
 732};
 733
 734#ifdef CONFIG_ANON_VMA_NAME
 735/*
 736 * mmap_lock should be read-locked when calling anon_vma_name(). Caller should
 737 * either keep holding the lock while using the returned pointer or it should
 738 * raise anon_vma_name refcount before releasing the lock.
 739 */
 740struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma);
 741struct anon_vma_name *anon_vma_name_alloc(const char *name);
 742void anon_vma_name_free(struct kref *kref);
 743#else /* CONFIG_ANON_VMA_NAME */
 744static inline struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma)
 745{
 746	return NULL;
 747}
 748
 749static inline struct anon_vma_name *anon_vma_name_alloc(const char *name)
 750{
 751	return NULL;
 752}
 753#endif
 754
 755#define VMA_LOCK_OFFSET	0x40000000
 756#define VMA_REF_LIMIT	(VMA_LOCK_OFFSET - 1)
 757
 758struct vma_numab_state {
 759	/*
 760	 * Initialised as time in 'jiffies' after which VMA
 761	 * should be scanned.  Delays first scan of new VMA by at
 762	 * least sysctl_numa_balancing_scan_delay:
 763	 */
 764	unsigned long next_scan;
 765
 766	/*
 767	 * Time in jiffies when pids_active[] is reset to
 768	 * detect phase change behaviour:
 769	 */
 770	unsigned long pids_active_reset;
 771
 772	/*
 773	 * Approximate tracking of PIDs that trapped a NUMA hinting
 774	 * fault. May produce false positives due to hash collisions.
 775	 *
 776	 *   [0] Previous PID tracking
 777	 *   [1] Current PID tracking
 778	 *
 779	 * Window moves after next_pid_reset has expired approximately
 780	 * every VMA_PID_RESET_PERIOD jiffies:
 781	 */
 782	unsigned long pids_active[2];
 783
 784	/* MM scan sequence ID when scan first started after VMA creation */
 785	int start_scan_seq;
 786
 787	/*
 788	 * MM scan sequence ID when the VMA was last completely scanned.
 789	 * A VMA is not eligible for scanning if prev_scan_seq == numa_scan_seq
 790	 */
 791	int prev_scan_seq;
 792};
 793
 794#ifdef __HAVE_PFNMAP_TRACKING
 795struct pfnmap_track_ctx {
 796	struct kref kref;
 797	unsigned long pfn;
 798	unsigned long size;	/* in bytes */
 799};
 800#endif
 801
 802/* What action should be taken after an .mmap_prepare call is complete? */
 803enum mmap_action_type {
 804	MMAP_NOTHING,		/* Mapping is complete, no further action. */
 805	MMAP_REMAP_PFN,		/* Remap PFN range. */
 806	MMAP_IO_REMAP_PFN,	/* I/O remap PFN range. */
 807};
 808
 809/*
 810 * Describes an action an mmap_prepare hook can instruct to be taken to complete
 811 * the mapping of a VMA. Specified in vm_area_desc.
 812 */
 813struct mmap_action {
 814	union {
 815		/* Remap range. */
 816		struct {
 817			unsigned long start;
 818			unsigned long start_pfn;
 819			unsigned long size;
 820			pgprot_t pgprot;
 821		} remap;
 822	};
 823	enum mmap_action_type type;
 824
 825	/*
 826	 * If specified, this hook is invoked after the selected action has been
 827	 * successfully completed. Note that the VMA write lock still held.
 828	 *
 829	 * The absolute minimum ought to be done here.
 830	 *
 831	 * Returns 0 on success, or an error code.
 832	 */
 833	int (*success_hook)(const struct vm_area_struct *vma);
 834
 835	/*
 836	 * If specified, this hook is invoked when an error occurred when
 837	 * attempting the selection action.
 838	 *
 839	 * The hook can return an error code in order to filter the error, but
 840	 * it is not valid to clear the error here.
 841	 */
 842	int (*error_hook)(int err);
 843
 844	/*
 845	 * This should be set in rare instances where the operation required
 846	 * that the rmap should not be able to access the VMA until
 847	 * completely set up.
 848	 */
 849	bool hide_from_rmap_until_complete :1;
 850};
 851
 852/*
 853 * Opaque type representing current VMA (vm_area_struct) flag state. Must be
 854 * accessed via vma_flags_xxx() helper functions.
 855 */
 856#define NUM_VMA_FLAG_BITS BITS_PER_LONG
 857typedef struct {
 858	DECLARE_BITMAP(__vma_flags, NUM_VMA_FLAG_BITS);
 859} __private vma_flags_t;
 860
 861/*
 862 * Describes a VMA that is about to be mmap()'ed. Drivers may choose to
 863 * manipulate mutable fields which will cause those fields to be updated in the
 864 * resultant VMA.
 865 *
 866 * Helper functions are not required for manipulating any field.
 867 */
 868struct vm_area_desc {
 869	/* Immutable state. */
 870	const struct mm_struct *const mm;
 871	struct file *const file; /* May vary from vm_file in stacked callers. */
 872	unsigned long start;
 873	unsigned long end;
 874
 875	/* Mutable fields. Populated with initial state. */
 876	pgoff_t pgoff;
 877	struct file *vm_file;
 878	union {
 879		vm_flags_t vm_flags;
 880		vma_flags_t vma_flags;
 881	};
 882	pgprot_t page_prot;
 883
 884	/* Write-only fields. */
 885	const struct vm_operations_struct *vm_ops;
 886	void *private_data;
 887
 888	/* Take further action? */
 889	struct mmap_action action;
 890};
 891
 892/*
 893 * This struct describes a virtual memory area. There is one of these
 894 * per VM-area/task. A VM area is any part of the process virtual memory
 895 * space that has a special rule for the page-fault handlers (ie a shared
 896 * library, the executable area etc).
 897 *
 898 * Only explicitly marked struct members may be accessed by RCU readers before
 899 * getting a stable reference.
 900 *
 901 * WARNING: when adding new members, please update vm_area_init_from() to copy
 902 * them during vm_area_struct content duplication.
 903 */
 904struct vm_area_struct {
 905	/* The first cache line has the info for VMA tree walking. */
 906
 907	union {
 908		struct {
 909			/* VMA covers [vm_start; vm_end) addresses within mm */
 910			unsigned long vm_start;
 911			unsigned long vm_end;
 912		};
 913		freeptr_t vm_freeptr; /* Pointer used by SLAB_TYPESAFE_BY_RCU */
 914	};
 915
 916	/*
 917	 * The address space we belong to.
 918	 * Unstable RCU readers are allowed to read this.
 919	 */
 920	struct mm_struct *vm_mm;
 921	pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 922
 923	/*
 924	 * Flags, see mm.h.
 925	 * To modify use vm_flags_{init|reset|set|clear|mod} functions.
 926	 * Preferably, use vma_flags_xxx() functions.
 927	 */
 928	union {
 929		/* Temporary while VMA flags are being converted. */
 930		const vm_flags_t vm_flags;
 931		vma_flags_t flags;
 932	};
 933
 934#ifdef CONFIG_PER_VMA_LOCK
 935	/*
 936	 * Can only be written (using WRITE_ONCE()) while holding both:
 937	 *  - mmap_lock (in write mode)
 938	 *  - vm_refcnt bit at VMA_LOCK_OFFSET is set
 939	 * Can be read reliably while holding one of:
 940	 *  - mmap_lock (in read or write mode)
 941	 *  - vm_refcnt bit at VMA_LOCK_OFFSET is set or vm_refcnt > 1
 942	 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
 943	 * while holding nothing (except RCU to keep the VMA struct allocated).
 944	 *
 945	 * This sequence counter is explicitly allowed to overflow; sequence
 946	 * counter reuse can only lead to occasional unnecessary use of the
 947	 * slowpath.
 948	 */
 949	unsigned int vm_lock_seq;
 950#endif
 951	/*
 952	 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 953	 * list, after a COW of one of the file pages.	A MAP_SHARED vma
 954	 * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 955	 * or brk vma (with NULL file) can only be in an anon_vma list.
 956	 */
 957	struct list_head anon_vma_chain; /* Serialized by mmap_lock &
 958					  * page_table_lock */
 959	struct anon_vma *anon_vma;	/* Serialized by page_table_lock */
 960
 961	/* Function pointers to deal with this struct. */
 962	const struct vm_operations_struct *vm_ops;
 963
 964	/* Information about our backing store: */
 965	unsigned long vm_pgoff;		/* Offset (within vm_file) in PAGE_SIZE
 966					   units */
 967	struct file * vm_file;		/* File we map to (can be NULL). */
 968	void * vm_private_data;		/* was vm_pte (shared mem) */
 969
 970#ifdef CONFIG_SWAP
 971	atomic_long_t swap_readahead_info;
 972#endif
 973#ifndef CONFIG_MMU
 974	struct vm_region *vm_region;	/* NOMMU mapping region */
 975#endif
 976#ifdef CONFIG_NUMA
 977	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
 978#endif
 979#ifdef CONFIG_NUMA_BALANCING
 980	struct vma_numab_state *numab_state;	/* NUMA Balancing state */
 981#endif
 982#ifdef CONFIG_PER_VMA_LOCK
 983	/* Unstable RCU readers are allowed to read this. */
 984	refcount_t vm_refcnt ____cacheline_aligned_in_smp;
 985#ifdef CONFIG_DEBUG_LOCK_ALLOC
 986	struct lockdep_map vmlock_dep_map;
 987#endif
 988#endif
 989	/*
 990	 * For areas with an address space and backing store,
 991	 * linkage into the address_space->i_mmap interval tree.
 992	 *
 993	 */
 994	struct {
 995		struct rb_node rb;
 996		unsigned long rb_subtree_last;
 997	} shared;
 998#ifdef CONFIG_ANON_VMA_NAME
 999	/*
1000	 * For private and shared anonymous mappings, a pointer to a null
1001	 * terminated string containing the name given to the vma, or NULL if
1002	 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access.
1003	 */
1004	struct anon_vma_name *anon_name;
1005#endif
1006	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
1007#ifdef __HAVE_PFNMAP_TRACKING
1008	struct pfnmap_track_ctx *pfnmap_track_ctx;
1009#endif
1010} __randomize_layout;
1011
1012/* Clears all bits in the VMA flags bitmap, non-atomically. */
1013static inline void vma_flags_clear_all(vma_flags_t *flags)
1014{
1015	bitmap_zero(ACCESS_PRIVATE(flags, __vma_flags), NUM_VMA_FLAG_BITS);
1016}
1017
1018/*
1019 * Copy value to the first system word of VMA flags, non-atomically.
1020 *
1021 * IMPORTANT: This does not overwrite bytes past the first system word. The
1022 * caller must account for this.
1023 */
1024static inline void vma_flags_overwrite_word(vma_flags_t *flags, unsigned long value)
1025{
1026	*ACCESS_PRIVATE(flags, __vma_flags) = value;
1027}
1028
1029/*
1030 * Copy value to the first system word of VMA flags ONCE, non-atomically.
1031 *
1032 * IMPORTANT: This does not overwrite bytes past the first system word. The
1033 * caller must account for this.
1034 */
1035static inline void vma_flags_overwrite_word_once(vma_flags_t *flags, unsigned long value)
1036{
1037	unsigned long *bitmap = ACCESS_PRIVATE(flags, __vma_flags);
1038
1039	WRITE_ONCE(*bitmap, value);
1040}
1041
1042/* Update the first system word of VMA flags setting bits, non-atomically. */
1043static inline void vma_flags_set_word(vma_flags_t *flags, unsigned long value)
1044{
1045	unsigned long *bitmap = ACCESS_PRIVATE(flags, __vma_flags);
1046
1047	*bitmap |= value;
1048}
1049
1050/* Update the first system word of VMA flags clearing bits, non-atomically. */
1051static inline void vma_flags_clear_word(vma_flags_t *flags, unsigned long value)
1052{
1053	unsigned long *bitmap = ACCESS_PRIVATE(flags, __vma_flags);
1054
1055	*bitmap &= ~value;
1056}
1057
1058#ifdef CONFIG_NUMA
1059#define vma_policy(vma) ((vma)->vm_policy)
1060#else
1061#define vma_policy(vma) NULL
1062#endif
1063
1064/*
1065 * Opaque type representing current mm_struct flag state. Must be accessed via
1066 * mm_flags_xxx() helper functions.
1067 */
1068#define NUM_MM_FLAG_BITS (64)
1069typedef struct {
1070	DECLARE_BITMAP(__mm_flags, NUM_MM_FLAG_BITS);
1071} __private mm_flags_t;
1072
1073struct kioctx_table;
1074struct iommu_mm_data;
1075struct mm_struct {
1076	struct {
1077		/*
1078		 * Fields which are often written to are placed in a separate
1079		 * cache line.
1080		 */
1081		struct {
1082			/**
1083			 * @mm_count: The number of references to &struct
1084			 * mm_struct (@mm_users count as 1).
1085			 *
1086			 * Use mmgrab()/mmdrop() to modify. When this drops to
1087			 * 0, the &struct mm_struct is freed.
1088			 */
1089			atomic_t mm_count;
1090		} ____cacheline_aligned_in_smp;
1091
1092		struct maple_tree mm_mt;
1093
1094		unsigned long mmap_base;	/* base of mmap area */
1095		unsigned long mmap_legacy_base;	/* base of mmap area in bottom-up allocations */
1096#ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
1097		/* Base addresses for compatible mmap() */
1098		unsigned long mmap_compat_base;
1099		unsigned long mmap_compat_legacy_base;
1100#endif
1101		unsigned long task_size;	/* size of task vm space */
1102		pgd_t * pgd;
1103
1104#ifdef CONFIG_MEMBARRIER
1105		/**
1106		 * @membarrier_state: Flags controlling membarrier behavior.
1107		 *
1108		 * This field is close to @pgd to hopefully fit in the same
1109		 * cache-line, which needs to be touched by switch_mm().
1110		 */
1111		atomic_t membarrier_state;
1112#endif
1113
1114		/**
1115		 * @mm_users: The number of users including userspace.
1116		 *
1117		 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
1118		 * drops to 0 (i.e. when the task exits and there are no other
1119		 * temporary reference holders), we also release a reference on
1120		 * @mm_count (which may then free the &struct mm_struct if
1121		 * @mm_count also drops to 0).
1122		 */
1123		atomic_t mm_users;
1124
1125		/* MM CID related storage */
1126		struct mm_mm_cid mm_cid;
1127
1128#ifdef CONFIG_MMU
1129		atomic_long_t pgtables_bytes;	/* size of all page tables */
1130#endif
1131		int map_count;			/* number of VMAs */
1132
1133		spinlock_t page_table_lock; /* Protects page tables and some
1134					     * counters
1135					     */
1136		/*
1137		 * Typically the current mmap_lock's offset is 56 bytes from
1138		 * the last cacheline boundary, which is very optimal, as
1139		 * its two hot fields 'count' and 'owner' sit in 2 different
1140		 * cachelines, and when mmap_lock is highly contended, both
1141		 * of the 2 fields will be accessed frequently, current layout
1142		 * will help to reduce cache bouncing.
1143		 *
1144		 * So please be careful with adding new fields before
1145		 * mmap_lock, which can easily push the 2 fields into one
1146		 * cacheline.
1147		 */
1148		struct rw_semaphore mmap_lock;
1149
1150		struct list_head mmlist; /* List of maybe swapped mm's.	These
1151					  * are globally strung together off
1152					  * init_mm.mmlist, and are protected
1153					  * by mmlist_lock
1154					  */
1155#ifdef CONFIG_PER_VMA_LOCK
1156		struct rcuwait vma_writer_wait;
1157		/*
1158		 * This field has lock-like semantics, meaning it is sometimes
1159		 * accessed with ACQUIRE/RELEASE semantics.
1160		 * Roughly speaking, incrementing the sequence number is
1161		 * equivalent to releasing locks on VMAs; reading the sequence
1162		 * number can be part of taking a read lock on a VMA.
1163		 * Incremented every time mmap_lock is write-locked/unlocked.
1164		 * Initialized to 0, therefore odd values indicate mmap_lock
1165		 * is write-locked and even values that it's released.
1166		 *
1167		 * Can be modified under write mmap_lock using RELEASE
1168		 * semantics.
1169		 * Can be read with no other protection when holding write
1170		 * mmap_lock.
1171		 * Can be read with ACQUIRE semantics if not holding write
1172		 * mmap_lock.
1173		 */
1174		seqcount_t mm_lock_seq;
1175#endif
1176#ifdef CONFIG_FUTEX_PRIVATE_HASH
1177		struct mutex			futex_hash_lock;
1178		struct futex_private_hash	__rcu *futex_phash;
1179		struct futex_private_hash	*futex_phash_new;
1180		/* futex-ref */
1181		unsigned long			futex_batches;
1182		struct rcu_head			futex_rcu;
1183		atomic_long_t			futex_atomic;
1184		unsigned int			__percpu *futex_ref;
1185#endif
1186
1187		unsigned long hiwater_rss; /* High-watermark of RSS usage */
1188		unsigned long hiwater_vm;  /* High-water virtual memory usage */
1189
1190		unsigned long total_vm;	   /* Total pages mapped */
1191		unsigned long locked_vm;   /* Pages that have PG_mlocked set */
1192		atomic64_t    pinned_vm;   /* Refcount permanently increased */
1193		unsigned long data_vm;	   /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
1194		unsigned long exec_vm;	   /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
1195		unsigned long stack_vm;	   /* VM_STACK */
1196		vm_flags_t def_flags;
1197
1198		/**
1199		 * @write_protect_seq: Locked when any thread is write
1200		 * protecting pages mapped by this mm to enforce a later COW,
1201		 * for instance during page table copying for fork().
1202		 */
1203		seqcount_t write_protect_seq;
1204
1205		spinlock_t arg_lock; /* protect the below fields */
1206
1207		unsigned long start_code, end_code, start_data, end_data;
1208		unsigned long start_brk, brk, start_stack;
1209		unsigned long arg_start, arg_end, env_start, env_end;
1210
1211		unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
1212
1213#ifdef CONFIG_ARCH_HAS_ELF_CORE_EFLAGS
1214		/* the ABI-related flags from the ELF header. Used for core dump */
1215		unsigned long saved_e_flags;
1216#endif
1217
1218		struct percpu_counter rss_stat[NR_MM_COUNTERS];
1219
1220		struct linux_binfmt *binfmt;
1221
1222		/* Architecture-specific MM context */
1223		mm_context_t context;
1224
1225		mm_flags_t flags; /* Must use mm_flags_* hlpers to access */
1226
1227#ifdef CONFIG_AIO
1228		spinlock_t			ioctx_lock;
1229		struct kioctx_table __rcu	*ioctx_table;
1230#endif
1231#ifdef CONFIG_MEMCG
1232		/*
1233		 * "owner" points to a task that is regarded as the canonical
1234		 * user/owner of this mm. All of the following must be true in
1235		 * order for it to be changed:
1236		 *
1237		 * current == mm->owner
1238		 * current->mm != mm
1239		 * new_owner->mm == mm
1240		 * new_owner->alloc_lock is held
1241		 */
1242		struct task_struct __rcu *owner;
1243#endif
1244		struct user_namespace *user_ns;
1245
1246		/* store ref to file /proc/<pid>/exe symlink points to */
1247		struct file __rcu *exe_file;
1248#ifdef CONFIG_MMU_NOTIFIER
1249		struct mmu_notifier_subscriptions *notifier_subscriptions;
1250#endif
1251#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
1252		pgtable_t pmd_huge_pte; /* protected by page_table_lock */
1253#endif
1254#ifdef CONFIG_NUMA_BALANCING
1255		/*
1256		 * numa_next_scan is the next time that PTEs will be remapped
1257		 * PROT_NONE to trigger NUMA hinting faults; such faults gather
1258		 * statistics and migrate pages to new nodes if necessary.
1259		 */
1260		unsigned long numa_next_scan;
1261
1262		/* Restart point for scanning and remapping PTEs. */
1263		unsigned long numa_scan_offset;
1264
1265		/* numa_scan_seq prevents two threads remapping PTEs. */
1266		int numa_scan_seq;
1267#endif
1268		/*
1269		 * An operation with batched TLB flushing is going on. Anything
1270		 * that can move process memory needs to flush the TLB when
1271		 * moving a PROT_NONE mapped page.
1272		 */
1273		atomic_t tlb_flush_pending;
1274#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
1275		/* See flush_tlb_batched_pending() */
1276		atomic_t tlb_flush_batched;
1277#endif
1278		struct uprobes_state uprobes_state;
1279#ifdef CONFIG_PREEMPT_RT
1280		struct rcu_head delayed_drop;
1281#endif
1282#ifdef CONFIG_HUGETLB_PAGE
1283		atomic_long_t hugetlb_usage;
1284#endif
1285		struct work_struct async_put_work;
1286
1287#ifdef CONFIG_IOMMU_MM_DATA
1288		struct iommu_mm_data *iommu_mm;
1289#endif
1290#ifdef CONFIG_KSM
1291		/*
1292		 * Represent how many pages of this process are involved in KSM
1293		 * merging (not including ksm_zero_pages).
1294		 */
1295		unsigned long ksm_merging_pages;
1296		/*
1297		 * Represent how many pages are checked for ksm merging
1298		 * including merged and not merged.
1299		 */
1300		unsigned long ksm_rmap_items;
1301		/*
1302		 * Represent how many empty pages are merged with kernel zero
1303		 * pages when enabling KSM use_zero_pages.
1304		 */
1305		atomic_long_t ksm_zero_pages;
1306#endif /* CONFIG_KSM */
1307#ifdef CONFIG_LRU_GEN_WALKS_MMU
1308		struct {
1309			/* this mm_struct is on lru_gen_mm_list */
1310			struct list_head list;
1311			/*
1312			 * Set when switching to this mm_struct, as a hint of
1313			 * whether it has been used since the last time per-node
1314			 * page table walkers cleared the corresponding bits.
1315			 */
1316			unsigned long bitmap;
1317#ifdef CONFIG_MEMCG
1318			/* points to the memcg of "owner" above */
1319			struct mem_cgroup *memcg;
1320#endif
1321		} lru_gen;
1322#endif /* CONFIG_LRU_GEN_WALKS_MMU */
1323#ifdef CONFIG_MM_ID
1324		mm_id_t mm_id;
1325#endif /* CONFIG_MM_ID */
1326	} __randomize_layout;
1327
1328	/*
1329	 * The mm_cpumask needs to be at the end of mm_struct, because it
1330	 * is dynamically sized based on nr_cpu_ids.
1331	 */
1332	char flexible_array[] __aligned(__alignof__(unsigned long));
1333};
1334
1335/* Copy value to the first system word of mm flags, non-atomically. */
1336static inline void __mm_flags_overwrite_word(struct mm_struct *mm, unsigned long value)
1337{
1338	*ACCESS_PRIVATE(&mm->flags, __mm_flags) = value;
1339}
1340
1341/* Obtain a read-only view of the mm flags bitmap. */
1342static inline const unsigned long *__mm_flags_get_bitmap(const struct mm_struct *mm)
1343{
1344	return (const unsigned long *)ACCESS_PRIVATE(&mm->flags, __mm_flags);
1345}
1346
1347/* Read the first system word of mm flags, non-atomically. */
1348static inline unsigned long __mm_flags_get_word(const struct mm_struct *mm)
1349{
1350	return *__mm_flags_get_bitmap(mm);
1351}
1352
1353/*
1354 * Update the first system word of mm flags ONLY, applying the specified mask to
1355 * it, then setting all flags specified by bits.
1356 */
1357static inline void __mm_flags_set_mask_bits_word(struct mm_struct *mm,
1358		unsigned long mask, unsigned long bits)
1359{
1360	unsigned long *bitmap = ACCESS_PRIVATE(&mm->flags, __mm_flags);
1361
1362	set_mask_bits(bitmap, mask, bits);
1363}
1364
1365#define MM_MT_FLAGS	(MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN | \
1366			 MT_FLAGS_USE_RCU)
1367extern struct mm_struct init_mm;
1368
1369#define MM_STRUCT_FLEXIBLE_ARRAY_INIT				\
1370{								\
1371	[0 ... sizeof(cpumask_t) + MM_CID_STATIC_SIZE - 1] = 0	\
1372}
1373
1374/* Pointer magic because the dynamic array size confuses some compilers. */
1375static inline void mm_init_cpumask(struct mm_struct *mm)
1376{
1377	unsigned long cpu_bitmap = (unsigned long)mm;
1378
1379	cpu_bitmap += offsetof(struct mm_struct, flexible_array);
1380	cpumask_clear((struct cpumask *)cpu_bitmap);
1381}
1382
1383/* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
1384static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
1385{
1386	return (struct cpumask *)&mm->flexible_array;
1387}
1388
1389#ifdef CONFIG_LRU_GEN
1390
1391struct lru_gen_mm_list {
1392	/* mm_struct list for page table walkers */
1393	struct list_head fifo;
1394	/* protects the list above */
1395	spinlock_t lock;
1396};
1397
1398#endif /* CONFIG_LRU_GEN */
1399
1400#ifdef CONFIG_LRU_GEN_WALKS_MMU
1401
1402void lru_gen_add_mm(struct mm_struct *mm);
1403void lru_gen_del_mm(struct mm_struct *mm);
1404void lru_gen_migrate_mm(struct mm_struct *mm);
1405
1406static inline void lru_gen_init_mm(struct mm_struct *mm)
1407{
1408	INIT_LIST_HEAD(&mm->lru_gen.list);
1409	mm->lru_gen.bitmap = 0;
1410#ifdef CONFIG_MEMCG
1411	mm->lru_gen.memcg = NULL;
1412#endif
1413}
1414
1415static inline void lru_gen_use_mm(struct mm_struct *mm)
1416{
1417	/*
1418	 * When the bitmap is set, page reclaim knows this mm_struct has been
1419	 * used since the last time it cleared the bitmap. So it might be worth
1420	 * walking the page tables of this mm_struct to clear the accessed bit.
1421	 */
1422	WRITE_ONCE(mm->lru_gen.bitmap, -1);
1423}
1424
1425#else /* !CONFIG_LRU_GEN_WALKS_MMU */
1426
1427static inline void lru_gen_add_mm(struct mm_struct *mm)
1428{
1429}
1430
1431static inline void lru_gen_del_mm(struct mm_struct *mm)
1432{
1433}
1434
1435static inline void lru_gen_migrate_mm(struct mm_struct *mm)
1436{
1437}
1438
1439static inline void lru_gen_init_mm(struct mm_struct *mm)
1440{
1441}
1442
1443static inline void lru_gen_use_mm(struct mm_struct *mm)
1444{
1445}
1446
1447#endif /* CONFIG_LRU_GEN_WALKS_MMU */
1448
1449struct vma_iterator {
1450	struct ma_state mas;
1451};
1452
1453#define VMA_ITERATOR(name, __mm, __addr)				\
1454	struct vma_iterator name = {					\
1455		.mas = {						\
1456			.tree = &(__mm)->mm_mt,				\
1457			.index = __addr,				\
1458			.node = NULL,					\
1459			.status = ma_start,				\
1460		},							\
1461	}
1462
1463static inline void vma_iter_init(struct vma_iterator *vmi,
1464		struct mm_struct *mm, unsigned long addr)
1465{
1466	mas_init(&vmi->mas, &mm->mm_mt, addr);
1467}
1468
1469#ifdef CONFIG_SCHED_MM_CID
1470/*
1471 * mm_cpus_allowed: Union of all mm's threads allowed CPUs.
1472 */
1473static inline cpumask_t *mm_cpus_allowed(struct mm_struct *mm)
1474{
1475	unsigned long bitmap = (unsigned long)mm;
1476
1477	bitmap += offsetof(struct mm_struct, flexible_array);
1478	/* Skip cpu_bitmap */
1479	bitmap += cpumask_size();
1480	return (struct cpumask *)bitmap;
1481}
1482
1483/* Accessor for struct mm_struct's cidmask. */
1484static inline unsigned long *mm_cidmask(struct mm_struct *mm)
1485{
1486	unsigned long cid_bitmap = (unsigned long)mm_cpus_allowed(mm);
1487
1488	/* Skip mm_cpus_allowed */
1489	cid_bitmap += cpumask_size();
1490	return (unsigned long *)cid_bitmap;
1491}
1492
1493void mm_init_cid(struct mm_struct *mm, struct task_struct *p);
1494
1495static inline int mm_alloc_cid_noprof(struct mm_struct *mm, struct task_struct *p)
1496{
1497	mm->mm_cid.pcpu = alloc_percpu_noprof(struct mm_cid_pcpu);
1498	if (!mm->mm_cid.pcpu)
1499		return -ENOMEM;
1500	mm_init_cid(mm, p);
1501	return 0;
1502}
1503# define mm_alloc_cid(...)	alloc_hooks(mm_alloc_cid_noprof(__VA_ARGS__))
1504
1505static inline void mm_destroy_cid(struct mm_struct *mm)
1506{
1507	free_percpu(mm->mm_cid.pcpu);
1508	mm->mm_cid.pcpu = NULL;
1509}
1510
1511static inline unsigned int mm_cid_size(void)
1512{
1513	/* mm_cpus_allowed(), mm_cidmask(). */
1514	return cpumask_size() + bitmap_size(num_possible_cpus());
1515}
1516
1517/* Use 2 * NR_CPUS as worse case for static allocation. */
1518# define MM_CID_STATIC_SIZE	(2 * sizeof(cpumask_t))
1519#else /* CONFIG_SCHED_MM_CID */
1520static inline void mm_init_cid(struct mm_struct *mm, struct task_struct *p) { }
1521static inline int mm_alloc_cid(struct mm_struct *mm, struct task_struct *p) { return 0; }
1522static inline void mm_destroy_cid(struct mm_struct *mm) { }
1523static inline unsigned int mm_cid_size(void)
1524{
1525	return 0;
1526}
1527# define MM_CID_STATIC_SIZE	0
1528#endif /* CONFIG_SCHED_MM_CID */
1529
1530struct mmu_gather;
1531extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
1532extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
1533void tlb_gather_mmu_vma(struct mmu_gather *tlb, struct vm_area_struct *vma);
1534extern void tlb_finish_mmu(struct mmu_gather *tlb);
1535
1536struct vm_fault;
1537
1538/**
1539 * typedef vm_fault_t - Return type for page fault handlers.
1540 *
1541 * Page fault handlers return a bitmask of %VM_FAULT values.
1542 */
1543typedef __bitwise unsigned int vm_fault_t;
1544
1545/**
1546 * enum vm_fault_reason - Page fault handlers return a bitmask of
1547 * these values to tell the core VM what happened when handling the
1548 * fault. Used to decide whether a process gets delivered SIGBUS or
1549 * just gets major/minor fault counters bumped up.
1550 *
1551 * @VM_FAULT_OOM:		Out Of Memory
1552 * @VM_FAULT_SIGBUS:		Bad access
1553 * @VM_FAULT_MAJOR:		Page read from storage
1554 * @VM_FAULT_HWPOISON:		Hit poisoned small page
1555 * @VM_FAULT_HWPOISON_LARGE:	Hit poisoned large page. Index encoded
1556 *				in upper bits
1557 * @VM_FAULT_SIGSEGV:		segmentation fault
1558 * @VM_FAULT_NOPAGE:		->fault installed the pte, not return page
1559 * @VM_FAULT_LOCKED:		->fault locked the returned page
1560 * @VM_FAULT_RETRY:		->fault blocked, must retry
1561 * @VM_FAULT_FALLBACK:		huge page fault failed, fall back to small
1562 * @VM_FAULT_DONE_COW:		->fault has fully handled COW
1563 * @VM_FAULT_NEEDDSYNC:		->fault did not modify page tables and needs
1564 *				fsync() to complete (for synchronous page faults
1565 *				in DAX)
1566 * @VM_FAULT_COMPLETED:		->fault completed, meanwhile mmap lock released
1567 * @VM_FAULT_HINDEX_MASK:	mask HINDEX value
1568 *
1569 */
1570enum vm_fault_reason {
1571	VM_FAULT_OOM            = (__force vm_fault_t)0x000001,
1572	VM_FAULT_SIGBUS         = (__force vm_fault_t)0x000002,
1573	VM_FAULT_MAJOR          = (__force vm_fault_t)0x000004,
1574	VM_FAULT_HWPOISON       = (__force vm_fault_t)0x000010,
1575	VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
1576	VM_FAULT_SIGSEGV        = (__force vm_fault_t)0x000040,
1577	VM_FAULT_NOPAGE         = (__force vm_fault_t)0x000100,
1578	VM_FAULT_LOCKED         = (__force vm_fault_t)0x000200,
1579	VM_FAULT_RETRY          = (__force vm_fault_t)0x000400,
1580	VM_FAULT_FALLBACK       = (__force vm_fault_t)0x000800,
1581	VM_FAULT_DONE_COW       = (__force vm_fault_t)0x001000,
1582	VM_FAULT_NEEDDSYNC      = (__force vm_fault_t)0x002000,
1583	VM_FAULT_COMPLETED      = (__force vm_fault_t)0x004000,
1584	VM_FAULT_HINDEX_MASK    = (__force vm_fault_t)0x0f0000,
1585};
1586
1587/* Encode hstate index for a hwpoisoned large page */
1588#define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
1589#define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
1590
1591#define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS |	\
1592			VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON |	\
1593			VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
1594
1595#define VM_FAULT_RESULT_TRACE \
1596	{ VM_FAULT_OOM,                 "OOM" },	\
1597	{ VM_FAULT_SIGBUS,              "SIGBUS" },	\
1598	{ VM_FAULT_MAJOR,               "MAJOR" },	\
1599	{ VM_FAULT_HWPOISON,            "HWPOISON" },	\
1600	{ VM_FAULT_HWPOISON_LARGE,      "HWPOISON_LARGE" },	\
1601	{ VM_FAULT_SIGSEGV,             "SIGSEGV" },	\
1602	{ VM_FAULT_NOPAGE,              "NOPAGE" },	\
1603	{ VM_FAULT_LOCKED,              "LOCKED" },	\
1604	{ VM_FAULT_RETRY,               "RETRY" },	\
1605	{ VM_FAULT_FALLBACK,            "FALLBACK" },	\
1606	{ VM_FAULT_DONE_COW,            "DONE_COW" },	\
1607	{ VM_FAULT_NEEDDSYNC,           "NEEDDSYNC" },	\
1608	{ VM_FAULT_COMPLETED,           "COMPLETED" }
1609
1610struct vm_special_mapping {
1611	const char *name;	/* The name, e.g. "[vdso]". */
1612
1613	/*
1614	 * If .fault is not provided, this points to a
1615	 * NULL-terminated array of pages that back the special mapping.
1616	 *
1617	 * This must not be NULL unless .fault is provided.
1618	 */
1619	struct page **pages;
1620
1621	/*
1622	 * If non-NULL, then this is called to resolve page faults
1623	 * on the special mapping.  If used, .pages is not checked.
1624	 */
1625	vm_fault_t (*fault)(const struct vm_special_mapping *sm,
1626				struct vm_area_struct *vma,
1627				struct vm_fault *vmf);
1628
1629	int (*mremap)(const struct vm_special_mapping *sm,
1630		     struct vm_area_struct *new_vma);
1631
1632	void (*close)(const struct vm_special_mapping *sm,
1633		      struct vm_area_struct *vma);
1634};
1635
1636enum tlb_flush_reason {
1637	TLB_FLUSH_ON_TASK_SWITCH,
1638	TLB_REMOTE_SHOOTDOWN,
1639	TLB_LOCAL_SHOOTDOWN,
1640	TLB_LOCAL_MM_SHOOTDOWN,
1641	TLB_REMOTE_SEND_IPI,
1642	TLB_REMOTE_WRONG_CPU,
1643};
1644
1645/**
1646 * enum fault_flag - Fault flag definitions.
1647 * @FAULT_FLAG_WRITE: Fault was a write fault.
1648 * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
1649 * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
1650 * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
1651 * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
1652 * @FAULT_FLAG_TRIED: The fault has been tried once.
1653 * @FAULT_FLAG_USER: The fault originated in userspace.
1654 * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
1655 * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
1656 * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
1657 * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a
1658 *                      COW mapping, making sure that an exclusive anon page is
1659 *                      mapped after the fault.
1660 * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached.
1661 *                        We should only access orig_pte if this flag set.
1662 * @FAULT_FLAG_VMA_LOCK: The fault is handled under VMA lock.
1663 *
1664 * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
1665 * whether we would allow page faults to retry by specifying these two
1666 * fault flags correctly.  Currently there can be three legal combinations:
1667 *
1668 * (a) ALLOW_RETRY and !TRIED:  this means the page fault allows retry, and
1669 *                              this is the first try
1670 *
1671 * (b) ALLOW_RETRY and TRIED:   this means the page fault allows retry, and
1672 *                              we've already tried at least once
1673 *
1674 * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
1675 *
1676 * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
1677 * be used.  Note that page faults can be allowed to retry for multiple times,
1678 * in which case we'll have an initial fault with flags (a) then later on
1679 * continuous faults with flags (b).  We should always try to detect pending
1680 * signals before a retry to make sure the continuous page faults can still be
1681 * interrupted if necessary.
1682 *
1683 * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal.
1684 * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when
1685 * applied to mappings that are not COW mappings.
1686 */
1687enum fault_flag {
1688	FAULT_FLAG_WRITE =		1 << 0,
1689	FAULT_FLAG_MKWRITE =		1 << 1,
1690	FAULT_FLAG_ALLOW_RETRY =	1 << 2,
1691	FAULT_FLAG_RETRY_NOWAIT = 	1 << 3,
1692	FAULT_FLAG_KILLABLE =		1 << 4,
1693	FAULT_FLAG_TRIED = 		1 << 5,
1694	FAULT_FLAG_USER =		1 << 6,
1695	FAULT_FLAG_REMOTE =		1 << 7,
1696	FAULT_FLAG_INSTRUCTION =	1 << 8,
1697	FAULT_FLAG_INTERRUPTIBLE =	1 << 9,
1698	FAULT_FLAG_UNSHARE =		1 << 10,
1699	FAULT_FLAG_ORIG_PTE_VALID =	1 << 11,
1700	FAULT_FLAG_VMA_LOCK =		1 << 12,
1701};
1702
1703typedef unsigned int __bitwise zap_flags_t;
1704
1705/* Flags for clear_young_dirty_ptes(). */
1706typedef int __bitwise cydp_t;
1707
1708/* Clear the access bit */
1709#define CYDP_CLEAR_YOUNG		((__force cydp_t)BIT(0))
1710
1711/* Clear the dirty bit */
1712#define CYDP_CLEAR_DIRTY		((__force cydp_t)BIT(1))
1713
1714/*
1715 * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each
1716 * other. Here is what they mean, and how to use them:
1717 *
1718 *
1719 * FIXME: For pages which are part of a filesystem, mappings are subject to the
1720 * lifetime enforced by the filesystem and we need guarantees that longterm
1721 * users like RDMA and V4L2 only establish mappings which coordinate usage with
1722 * the filesystem.  Ideas for this coordination include revoking the longterm
1723 * pin, delaying writeback, bounce buffer page writeback, etc.  As FS DAX was
1724 * added after the problem with filesystems was found FS DAX VMAs are
1725 * specifically failed.  Filesystem pages are still subject to bugs and use of
1726 * FOLL_LONGTERM should be avoided on those pages.
1727 *
1728 * In the CMA case: long term pins in a CMA region would unnecessarily fragment
1729 * that region.  And so, CMA attempts to migrate the page before pinning, when
1730 * FOLL_LONGTERM is specified.
1731 *
1732 * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount,
1733 * but an additional pin counting system) will be invoked. This is intended for
1734 * anything that gets a page reference and then touches page data (for example,
1735 * Direct IO). This lets the filesystem know that some non-file-system entity is
1736 * potentially changing the pages' data. In contrast to FOLL_GET (whose pages
1737 * are released via put_page()), FOLL_PIN pages must be released, ultimately, by
1738 * a call to unpin_user_page().
1739 *
1740 * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different
1741 * and separate refcounting mechanisms, however, and that means that each has
1742 * its own acquire and release mechanisms:
1743 *
1744 *     FOLL_GET: get_user_pages*() to acquire, and put_page() to release.
1745 *
1746 *     FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release.
1747 *
1748 * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call.
1749 * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based
1750 * calls applied to them, and that's perfectly OK. This is a constraint on the
1751 * callers, not on the pages.)
1752 *
1753 * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never
1754 * directly by the caller. That's in order to help avoid mismatches when
1755 * releasing pages: get_user_pages*() pages must be released via put_page(),
1756 * while pin_user_pages*() pages must be released via unpin_user_page().
1757 *
1758 * Please see Documentation/core-api/pin_user_pages.rst for more information.
1759 */
1760
1761enum {
1762	/* check pte is writable */
1763	FOLL_WRITE = 1 << 0,
1764	/* do get_page on page */
1765	FOLL_GET = 1 << 1,
1766	/* give error on hole if it would be zero */
1767	FOLL_DUMP = 1 << 2,
1768	/* get_user_pages read/write w/o permission */
1769	FOLL_FORCE = 1 << 3,
1770	/*
1771	 * if a disk transfer is needed, start the IO and return without waiting
1772	 * upon it
1773	 */
1774	FOLL_NOWAIT = 1 << 4,
1775	/* do not fault in pages */
1776	FOLL_NOFAULT = 1 << 5,
1777	/* check page is hwpoisoned */
1778	FOLL_HWPOISON = 1 << 6,
1779	/* don't do file mappings */
1780	FOLL_ANON = 1 << 7,
1781	/*
1782	 * FOLL_LONGTERM indicates that the page will be held for an indefinite
1783	 * time period _often_ under userspace control.  This is in contrast to
1784	 * iov_iter_get_pages(), whose usages are transient.
1785	 */
1786	FOLL_LONGTERM = 1 << 8,
1787	/* split huge pmd before returning */
1788	FOLL_SPLIT_PMD = 1 << 9,
1789	/* allow returning PCI P2PDMA pages */
1790	FOLL_PCI_P2PDMA = 1 << 10,
1791	/* allow interrupts from generic signals */
1792	FOLL_INTERRUPTIBLE = 1 << 11,
1793	/*
1794	 * Always honor (trigger) NUMA hinting faults.
1795	 *
1796	 * FOLL_WRITE implicitly honors NUMA hinting faults because a
1797	 * PROT_NONE-mapped page is not writable (exceptions with FOLL_FORCE
1798	 * apply). get_user_pages_fast_only() always implicitly honors NUMA
1799	 * hinting faults.
1800	 */
1801	FOLL_HONOR_NUMA_FAULT = 1 << 12,
1802
1803	/* See also internal only FOLL flags in mm/internal.h */
1804};
1805
1806/* mm flags */
1807
1808/*
1809 * The first two bits represent core dump modes for set-user-ID,
1810 * the modes are SUID_DUMP_* defined in linux/sched/coredump.h
1811 */
1812#define MMF_DUMPABLE_BITS 2
1813#define MMF_DUMPABLE_MASK (BIT(MMF_DUMPABLE_BITS) - 1)
1814/* coredump filter bits */
1815#define MMF_DUMP_ANON_PRIVATE	2
1816#define MMF_DUMP_ANON_SHARED	3
1817#define MMF_DUMP_MAPPED_PRIVATE	4
1818#define MMF_DUMP_MAPPED_SHARED	5
1819#define MMF_DUMP_ELF_HEADERS	6
1820#define MMF_DUMP_HUGETLB_PRIVATE 7
1821#define MMF_DUMP_HUGETLB_SHARED  8
1822#define MMF_DUMP_DAX_PRIVATE	9
1823#define MMF_DUMP_DAX_SHARED	10
1824
1825#define MMF_DUMP_FILTER_SHIFT	MMF_DUMPABLE_BITS
1826#define MMF_DUMP_FILTER_BITS	9
1827#define MMF_DUMP_FILTER_MASK \
1828	((BIT(MMF_DUMP_FILTER_BITS) - 1) << MMF_DUMP_FILTER_SHIFT)
1829#define MMF_DUMP_FILTER_DEFAULT \
1830	(BIT(MMF_DUMP_ANON_PRIVATE) | BIT(MMF_DUMP_ANON_SHARED) | \
1831	 BIT(MMF_DUMP_HUGETLB_PRIVATE) | MMF_DUMP_MASK_DEFAULT_ELF)
1832
1833#ifdef CONFIG_CORE_DUMP_DEFAULT_ELF_HEADERS
1834# define MMF_DUMP_MASK_DEFAULT_ELF	BIT(MMF_DUMP_ELF_HEADERS)
1835#else
1836# define MMF_DUMP_MASK_DEFAULT_ELF	0
1837#endif
1838					/* leave room for more dump flags */
1839#define MMF_VM_MERGEABLE	16	/* KSM may merge identical pages */
1840#define MMF_VM_HUGEPAGE		17	/* set when mm is available for khugepaged */
1841
1842#define MMF_HUGE_ZERO_FOLIO	18      /* mm has ever used the global huge zero folio */
1843
1844#define MMF_HAS_UPROBES		19	/* has uprobes */
1845#define MMF_RECALC_UPROBES	20	/* MMF_HAS_UPROBES can be wrong */
1846#define MMF_OOM_SKIP		21	/* mm is of no interest for the OOM killer */
1847#define MMF_UNSTABLE		22	/* mm is unstable for copy_from_user */
1848#define MMF_DISABLE_THP_EXCEPT_ADVISED	23	/* no THP except when advised (e.g., VM_HUGEPAGE) */
1849#define MMF_DISABLE_THP_COMPLETELY	24	/* no THP for all VMAs */
1850#define MMF_DISABLE_THP_MASK	(BIT(MMF_DISABLE_THP_COMPLETELY) | \
1851				 BIT(MMF_DISABLE_THP_EXCEPT_ADVISED))
1852#define MMF_OOM_REAP_QUEUED	25	/* mm was queued for oom_reaper */
1853#define MMF_MULTIPROCESS	26	/* mm is shared between processes */
1854/*
1855 * MMF_HAS_PINNED: Whether this mm has pinned any pages.  This can be either
1856 * replaced in the future by mm.pinned_vm when it becomes stable, or grow into
1857 * a counter on its own. We're aggresive on this bit for now: even if the
1858 * pinned pages were unpinned later on, we'll still keep this bit set for the
1859 * lifecycle of this mm, just for simplicity.
1860 */
1861#define MMF_HAS_PINNED		27	/* FOLL_PIN has run, never cleared */
1862
1863#define MMF_HAS_MDWE		28
1864#define MMF_HAS_MDWE_MASK	BIT(MMF_HAS_MDWE)
1865
1866#define MMF_HAS_MDWE_NO_INHERIT	29
1867
1868#define MMF_VM_MERGE_ANY	30
1869#define MMF_VM_MERGE_ANY_MASK	BIT(MMF_VM_MERGE_ANY)
1870
1871#define MMF_TOPDOWN		31	/* mm searches top down by default */
1872#define MMF_TOPDOWN_MASK	BIT(MMF_TOPDOWN)
1873
1874#define MMF_INIT_LEGACY_MASK	(MMF_DUMPABLE_MASK | MMF_DUMP_FILTER_MASK |\
1875				 MMF_DISABLE_THP_MASK | MMF_HAS_MDWE_MASK |\
1876				 MMF_VM_MERGE_ANY_MASK | MMF_TOPDOWN_MASK)
1877
1878/* Legacy flags must fit within 32 bits. */
1879static_assert((u64)MMF_INIT_LEGACY_MASK <= (u64)UINT_MAX);
1880
1881/*
1882 * Initialise legacy flags according to masks, propagating selected flags on
1883 * fork. Further flag manipulation can be performed by the caller.
1884 */
1885static inline unsigned long mmf_init_legacy_flags(unsigned long flags)
1886{
1887	if (flags & (1UL << MMF_HAS_MDWE_NO_INHERIT))
1888		flags &= ~((1UL << MMF_HAS_MDWE) |
1889			   (1UL << MMF_HAS_MDWE_NO_INHERIT));
1890	return flags & MMF_INIT_LEGACY_MASK;
1891}
1892
1893#endif /* _LINUX_MM_TYPES_H */