xref: /linux-6.15/include/linux/mm_types.h (revision d0f482bb)
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/list.h>
9 #include <linux/spinlock.h>
10 #include <linux/rbtree.h>
11 #include <linux/rwsem.h>
12 #include <linux/completion.h>
13 #include <linux/cpumask.h>
14 #include <linux/uprobes.h>
15 #include <linux/page-flags-layout.h>
16 #include <linux/workqueue.h>
17 #include <linux/seqlock.h>
18 
19 #include <asm/mmu.h>
20 
21 #ifndef AT_VECTOR_SIZE_ARCH
22 #define AT_VECTOR_SIZE_ARCH 0
23 #endif
24 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
25 
26 #define INIT_PASID	0
27 
28 struct address_space;
29 struct mem_cgroup;
30 
31 /*
32  * Each physical page in the system has a struct page associated with
33  * it to keep track of whatever it is we are using the page for at the
34  * moment. Note that we have no way to track which tasks are using
35  * a page, though if it is a pagecache page, rmap structures can tell us
36  * who is mapping it.
37  *
38  * If you allocate the page using alloc_pages(), you can use some of the
39  * space in struct page for your own purposes.  The five words in the main
40  * union are available, except for bit 0 of the first word which must be
41  * kept clear.  Many users use this word to store a pointer to an object
42  * which is guaranteed to be aligned.  If you use the same storage as
43  * page->mapping, you must restore it to NULL before freeing the page.
44  *
45  * If your page will not be mapped to userspace, you can also use the four
46  * bytes in the mapcount union, but you must call page_mapcount_reset()
47  * before freeing it.
48  *
49  * If you want to use the refcount field, it must be used in such a way
50  * that other CPUs temporarily incrementing and then decrementing the
51  * refcount does not cause problems.  On receiving the page from
52  * alloc_pages(), the refcount will be positive.
53  *
54  * If you allocate pages of order > 0, you can use some of the fields
55  * in each subpage, but you may need to restore some of their values
56  * afterwards.
57  *
58  * SLUB uses cmpxchg_double() to atomically update its freelist and
59  * counters.  That requires that freelist & counters be adjacent and
60  * double-word aligned.  We align all struct pages to double-word
61  * boundaries, and ensure that 'freelist' is aligned within the
62  * struct.
63  */
64 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
65 #define _struct_page_alignment	__aligned(2 * sizeof(unsigned long))
66 #else
67 #define _struct_page_alignment
68 #endif
69 
70 struct page {
71 	unsigned long flags;		/* Atomic flags, some possibly
72 					 * updated asynchronously */
73 	/*
74 	 * Five words (20/40 bytes) are available in this union.
75 	 * WARNING: bit 0 of the first word is used for PageTail(). That
76 	 * means the other users of this union MUST NOT use the bit to
77 	 * avoid collision and false-positive PageTail().
78 	 */
79 	union {
80 		struct {	/* Page cache and anonymous pages */
81 			/**
82 			 * @lru: Pageout list, eg. active_list protected by
83 			 * lruvec->lru_lock.  Sometimes used as a generic list
84 			 * by the page owner.
85 			 */
86 			struct list_head lru;
87 			/* See page-flags.h for PAGE_MAPPING_FLAGS */
88 			struct address_space *mapping;
89 			pgoff_t index;		/* Our offset within mapping. */
90 			/**
91 			 * @private: Mapping-private opaque data.
92 			 * Usually used for buffer_heads if PagePrivate.
93 			 * Used for swp_entry_t if PageSwapCache.
94 			 * Indicates order in the buddy system if PageBuddy.
95 			 */
96 			unsigned long private;
97 		};
98 		struct {	/* page_pool used by netstack */
99 			/**
100 			 * @pp_magic: magic value to avoid recycling non
101 			 * page_pool allocated pages.
102 			 */
103 			unsigned long pp_magic;
104 			struct page_pool *pp;
105 			unsigned long _pp_mapping_pad;
106 			/**
107 			 * @dma_addr: might require a 64-bit value on
108 			 * 32-bit architectures.
109 			 */
110 			unsigned long dma_addr[2];
111 		};
112 		struct {	/* slab, slob and slub */
113 			union {
114 				struct list_head slab_list;
115 				struct {	/* Partial pages */
116 					struct page *next;
117 #ifdef CONFIG_64BIT
118 					int pages;	/* Nr of pages left */
119 					int pobjects;	/* Approximate count */
120 #else
121 					short int pages;
122 					short int pobjects;
123 #endif
124 				};
125 			};
126 			struct kmem_cache *slab_cache; /* not slob */
127 			/* Double-word boundary */
128 			void *freelist;		/* first free object */
129 			union {
130 				void *s_mem;	/* slab: first object */
131 				unsigned long counters;		/* SLUB */
132 				struct {			/* SLUB */
133 					unsigned inuse:16;
134 					unsigned objects:15;
135 					unsigned frozen:1;
136 				};
137 			};
138 		};
139 		struct {	/* Tail pages of compound page */
140 			unsigned long compound_head;	/* Bit zero is set */
141 
142 			/* First tail page only */
143 			unsigned char compound_dtor;
144 			unsigned char compound_order;
145 			atomic_t compound_mapcount;
146 			unsigned int compound_nr; /* 1 << compound_order */
147 		};
148 		struct {	/* Second tail page of compound page */
149 			unsigned long _compound_pad_1;	/* compound_head */
150 			atomic_t hpage_pinned_refcount;
151 			/* For both global and memcg */
152 			struct list_head deferred_list;
153 		};
154 		struct {	/* Page table pages */
155 			unsigned long _pt_pad_1;	/* compound_head */
156 			pgtable_t pmd_huge_pte; /* protected by page->ptl */
157 			unsigned long _pt_pad_2;	/* mapping */
158 			union {
159 				struct mm_struct *pt_mm; /* x86 pgds only */
160 				atomic_t pt_frag_refcount; /* powerpc */
161 			};
162 #if ALLOC_SPLIT_PTLOCKS
163 			spinlock_t *ptl;
164 #else
165 			spinlock_t ptl;
166 #endif
167 		};
168 		struct {	/* ZONE_DEVICE pages */
169 			/** @pgmap: Points to the hosting device page map. */
170 			struct dev_pagemap *pgmap;
171 			void *zone_device_data;
172 			/*
173 			 * ZONE_DEVICE private pages are counted as being
174 			 * mapped so the next 3 words hold the mapping, index,
175 			 * and private fields from the source anonymous or
176 			 * page cache page while the page is migrated to device
177 			 * private memory.
178 			 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
179 			 * use the mapping, index, and private fields when
180 			 * pmem backed DAX files are mapped.
181 			 */
182 		};
183 
184 		/** @rcu_head: You can use this to free a page by RCU. */
185 		struct rcu_head rcu_head;
186 	};
187 
188 	union {		/* This union is 4 bytes in size. */
189 		/*
190 		 * If the page can be mapped to userspace, encodes the number
191 		 * of times this page is referenced by a page table.
192 		 */
193 		atomic_t _mapcount;
194 
195 		/*
196 		 * If the page is neither PageSlab nor mappable to userspace,
197 		 * the value stored here may help determine what this page
198 		 * is used for.  See page-flags.h for a list of page types
199 		 * which are currently stored here.
200 		 */
201 		unsigned int page_type;
202 
203 		unsigned int active;		/* SLAB */
204 		int units;			/* SLOB */
205 	};
206 
207 	/* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
208 	atomic_t _refcount;
209 
210 #ifdef CONFIG_MEMCG
211 	unsigned long memcg_data;
212 #endif
213 
214 	/*
215 	 * On machines where all RAM is mapped into kernel address space,
216 	 * we can simply calculate the virtual address. On machines with
217 	 * highmem some memory is mapped into kernel virtual memory
218 	 * dynamically, so we need a place to store that address.
219 	 * Note that this field could be 16 bits on x86 ... ;)
220 	 *
221 	 * Architectures with slow multiplication can define
222 	 * WANT_PAGE_VIRTUAL in asm/page.h
223 	 */
224 #if defined(WANT_PAGE_VIRTUAL)
225 	void *virtual;			/* Kernel virtual address (NULL if
226 					   not kmapped, ie. highmem) */
227 #endif /* WANT_PAGE_VIRTUAL */
228 
229 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
230 	int _last_cpupid;
231 #endif
232 } _struct_page_alignment;
233 
234 static inline atomic_t *compound_mapcount_ptr(struct page *page)
235 {
236 	return &page[1].compound_mapcount;
237 }
238 
239 static inline atomic_t *compound_pincount_ptr(struct page *page)
240 {
241 	return &page[2].hpage_pinned_refcount;
242 }
243 
244 /*
245  * Used for sizing the vmemmap region on some architectures
246  */
247 #define STRUCT_PAGE_MAX_SHIFT	(order_base_2(sizeof(struct page)))
248 
249 #define PAGE_FRAG_CACHE_MAX_SIZE	__ALIGN_MASK(32768, ~PAGE_MASK)
250 #define PAGE_FRAG_CACHE_MAX_ORDER	get_order(PAGE_FRAG_CACHE_MAX_SIZE)
251 
252 #define page_private(page)		((page)->private)
253 
254 static inline void set_page_private(struct page *page, unsigned long private)
255 {
256 	page->private = private;
257 }
258 
259 struct page_frag_cache {
260 	void * va;
261 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
262 	__u16 offset;
263 	__u16 size;
264 #else
265 	__u32 offset;
266 #endif
267 	/* we maintain a pagecount bias, so that we dont dirty cache line
268 	 * containing page->_refcount every time we allocate a fragment.
269 	 */
270 	unsigned int		pagecnt_bias;
271 	bool pfmemalloc;
272 };
273 
274 typedef unsigned long vm_flags_t;
275 
276 /*
277  * A region containing a mapping of a non-memory backed file under NOMMU
278  * conditions.  These are held in a global tree and are pinned by the VMAs that
279  * map parts of them.
280  */
281 struct vm_region {
282 	struct rb_node	vm_rb;		/* link in global region tree */
283 	vm_flags_t	vm_flags;	/* VMA vm_flags */
284 	unsigned long	vm_start;	/* start address of region */
285 	unsigned long	vm_end;		/* region initialised to here */
286 	unsigned long	vm_top;		/* region allocated to here */
287 	unsigned long	vm_pgoff;	/* the offset in vm_file corresponding to vm_start */
288 	struct file	*vm_file;	/* the backing file or NULL */
289 
290 	int		vm_usage;	/* region usage count (access under nommu_region_sem) */
291 	bool		vm_icache_flushed : 1; /* true if the icache has been flushed for
292 						* this region */
293 };
294 
295 #ifdef CONFIG_USERFAULTFD
296 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
297 struct vm_userfaultfd_ctx {
298 	struct userfaultfd_ctx *ctx;
299 };
300 #else /* CONFIG_USERFAULTFD */
301 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
302 struct vm_userfaultfd_ctx {};
303 #endif /* CONFIG_USERFAULTFD */
304 
305 /*
306  * This struct describes a virtual memory area. There is one of these
307  * per VM-area/task. A VM area is any part of the process virtual memory
308  * space that has a special rule for the page-fault handlers (ie a shared
309  * library, the executable area etc).
310  */
311 struct vm_area_struct {
312 	/* The first cache line has the info for VMA tree walking. */
313 
314 	unsigned long vm_start;		/* Our start address within vm_mm. */
315 	unsigned long vm_end;		/* The first byte after our end address
316 					   within vm_mm. */
317 
318 	/* linked list of VM areas per task, sorted by address */
319 	struct vm_area_struct *vm_next, *vm_prev;
320 
321 	struct rb_node vm_rb;
322 
323 	/*
324 	 * Largest free memory gap in bytes to the left of this VMA.
325 	 * Either between this VMA and vma->vm_prev, or between one of the
326 	 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
327 	 * get_unmapped_area find a free area of the right size.
328 	 */
329 	unsigned long rb_subtree_gap;
330 
331 	/* Second cache line starts here. */
332 
333 	struct mm_struct *vm_mm;	/* The address space we belong to. */
334 
335 	/*
336 	 * Access permissions of this VMA.
337 	 * See vmf_insert_mixed_prot() for discussion.
338 	 */
339 	pgprot_t vm_page_prot;
340 	unsigned long vm_flags;		/* Flags, see mm.h. */
341 
342 	/*
343 	 * For areas with an address space and backing store,
344 	 * linkage into the address_space->i_mmap interval tree.
345 	 */
346 	struct {
347 		struct rb_node rb;
348 		unsigned long rb_subtree_last;
349 	} shared;
350 
351 	/*
352 	 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
353 	 * list, after a COW of one of the file pages.	A MAP_SHARED vma
354 	 * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
355 	 * or brk vma (with NULL file) can only be in an anon_vma list.
356 	 */
357 	struct list_head anon_vma_chain; /* Serialized by mmap_lock &
358 					  * page_table_lock */
359 	struct anon_vma *anon_vma;	/* Serialized by page_table_lock */
360 
361 	/* Function pointers to deal with this struct. */
362 	const struct vm_operations_struct *vm_ops;
363 
364 	/* Information about our backing store: */
365 	unsigned long vm_pgoff;		/* Offset (within vm_file) in PAGE_SIZE
366 					   units */
367 	struct file * vm_file;		/* File we map to (can be NULL). */
368 	void * vm_private_data;		/* was vm_pte (shared mem) */
369 
370 #ifdef CONFIG_SWAP
371 	atomic_long_t swap_readahead_info;
372 #endif
373 #ifndef CONFIG_MMU
374 	struct vm_region *vm_region;	/* NOMMU mapping region */
375 #endif
376 #ifdef CONFIG_NUMA
377 	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
378 #endif
379 	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
380 } __randomize_layout;
381 
382 struct core_thread {
383 	struct task_struct *task;
384 	struct core_thread *next;
385 };
386 
387 struct core_state {
388 	atomic_t nr_threads;
389 	struct core_thread dumper;
390 	struct completion startup;
391 };
392 
393 struct kioctx_table;
394 struct mm_struct {
395 	struct {
396 		struct vm_area_struct *mmap;		/* list of VMAs */
397 		struct rb_root mm_rb;
398 		u64 vmacache_seqnum;                   /* per-thread vmacache */
399 #ifdef CONFIG_MMU
400 		unsigned long (*get_unmapped_area) (struct file *filp,
401 				unsigned long addr, unsigned long len,
402 				unsigned long pgoff, unsigned long flags);
403 #endif
404 		unsigned long mmap_base;	/* base of mmap area */
405 		unsigned long mmap_legacy_base;	/* base of mmap area in bottom-up allocations */
406 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
407 		/* Base adresses for compatible mmap() */
408 		unsigned long mmap_compat_base;
409 		unsigned long mmap_compat_legacy_base;
410 #endif
411 		unsigned long task_size;	/* size of task vm space */
412 		unsigned long highest_vm_end;	/* highest vma end address */
413 		pgd_t * pgd;
414 
415 #ifdef CONFIG_MEMBARRIER
416 		/**
417 		 * @membarrier_state: Flags controlling membarrier behavior.
418 		 *
419 		 * This field is close to @pgd to hopefully fit in the same
420 		 * cache-line, which needs to be touched by switch_mm().
421 		 */
422 		atomic_t membarrier_state;
423 #endif
424 
425 		/**
426 		 * @mm_users: The number of users including userspace.
427 		 *
428 		 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
429 		 * drops to 0 (i.e. when the task exits and there are no other
430 		 * temporary reference holders), we also release a reference on
431 		 * @mm_count (which may then free the &struct mm_struct if
432 		 * @mm_count also drops to 0).
433 		 */
434 		atomic_t mm_users;
435 
436 		/**
437 		 * @mm_count: The number of references to &struct mm_struct
438 		 * (@mm_users count as 1).
439 		 *
440 		 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
441 		 * &struct mm_struct is freed.
442 		 */
443 		atomic_t mm_count;
444 
445 		/**
446 		 * @has_pinned: Whether this mm has pinned any pages.  This can
447 		 * be either replaced in the future by @pinned_vm when it
448 		 * becomes stable, or grow into a counter on its own. We're
449 		 * aggresive on this bit now - even if the pinned pages were
450 		 * unpinned later on, we'll still keep this bit set for the
451 		 * lifecycle of this mm just for simplicity.
452 		 */
453 		atomic_t has_pinned;
454 
455 #ifdef CONFIG_MMU
456 		atomic_long_t pgtables_bytes;	/* PTE page table pages */
457 #endif
458 		int map_count;			/* number of VMAs */
459 
460 		spinlock_t page_table_lock; /* Protects page tables and some
461 					     * counters
462 					     */
463 		/*
464 		 * With some kernel config, the current mmap_lock's offset
465 		 * inside 'mm_struct' is at 0x120, which is very optimal, as
466 		 * its two hot fields 'count' and 'owner' sit in 2 different
467 		 * cachelines,  and when mmap_lock is highly contended, both
468 		 * of the 2 fields will be accessed frequently, current layout
469 		 * will help to reduce cache bouncing.
470 		 *
471 		 * So please be careful with adding new fields before
472 		 * mmap_lock, which can easily push the 2 fields into one
473 		 * cacheline.
474 		 */
475 		struct rw_semaphore mmap_lock;
476 
477 		struct list_head mmlist; /* List of maybe swapped mm's.	These
478 					  * are globally strung together off
479 					  * init_mm.mmlist, and are protected
480 					  * by mmlist_lock
481 					  */
482 
483 
484 		unsigned long hiwater_rss; /* High-watermark of RSS usage */
485 		unsigned long hiwater_vm;  /* High-water virtual memory usage */
486 
487 		unsigned long total_vm;	   /* Total pages mapped */
488 		unsigned long locked_vm;   /* Pages that have PG_mlocked set */
489 		atomic64_t    pinned_vm;   /* Refcount permanently increased */
490 		unsigned long data_vm;	   /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
491 		unsigned long exec_vm;	   /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
492 		unsigned long stack_vm;	   /* VM_STACK */
493 		unsigned long def_flags;
494 
495 		/**
496 		 * @write_protect_seq: Locked when any thread is write
497 		 * protecting pages mapped by this mm to enforce a later COW,
498 		 * for instance during page table copying for fork().
499 		 */
500 		seqcount_t write_protect_seq;
501 
502 		spinlock_t arg_lock; /* protect the below fields */
503 
504 		unsigned long start_code, end_code, start_data, end_data;
505 		unsigned long start_brk, brk, start_stack;
506 		unsigned long arg_start, arg_end, env_start, env_end;
507 
508 		unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
509 
510 		/*
511 		 * Special counters, in some configurations protected by the
512 		 * page_table_lock, in other configurations by being atomic.
513 		 */
514 		struct mm_rss_stat rss_stat;
515 
516 		struct linux_binfmt *binfmt;
517 
518 		/* Architecture-specific MM context */
519 		mm_context_t context;
520 
521 		unsigned long flags; /* Must use atomic bitops to access */
522 
523 		struct core_state *core_state; /* coredumping support */
524 
525 #ifdef CONFIG_AIO
526 		spinlock_t			ioctx_lock;
527 		struct kioctx_table __rcu	*ioctx_table;
528 #endif
529 #ifdef CONFIG_MEMCG
530 		/*
531 		 * "owner" points to a task that is regarded as the canonical
532 		 * user/owner of this mm. All of the following must be true in
533 		 * order for it to be changed:
534 		 *
535 		 * current == mm->owner
536 		 * current->mm != mm
537 		 * new_owner->mm == mm
538 		 * new_owner->alloc_lock is held
539 		 */
540 		struct task_struct __rcu *owner;
541 #endif
542 		struct user_namespace *user_ns;
543 
544 		/* store ref to file /proc/<pid>/exe symlink points to */
545 		struct file __rcu *exe_file;
546 #ifdef CONFIG_MMU_NOTIFIER
547 		struct mmu_notifier_subscriptions *notifier_subscriptions;
548 #endif
549 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
550 		pgtable_t pmd_huge_pte; /* protected by page_table_lock */
551 #endif
552 #ifdef CONFIG_NUMA_BALANCING
553 		/*
554 		 * numa_next_scan is the next time that the PTEs will be marked
555 		 * pte_numa. NUMA hinting faults will gather statistics and
556 		 * migrate pages to new nodes if necessary.
557 		 */
558 		unsigned long numa_next_scan;
559 
560 		/* Restart point for scanning and setting pte_numa */
561 		unsigned long numa_scan_offset;
562 
563 		/* numa_scan_seq prevents two threads setting pte_numa */
564 		int numa_scan_seq;
565 #endif
566 		/*
567 		 * An operation with batched TLB flushing is going on. Anything
568 		 * that can move process memory needs to flush the TLB when
569 		 * moving a PROT_NONE or PROT_NUMA mapped page.
570 		 */
571 		atomic_t tlb_flush_pending;
572 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
573 		/* See flush_tlb_batched_pending() */
574 		bool tlb_flush_batched;
575 #endif
576 		struct uprobes_state uprobes_state;
577 #ifdef CONFIG_HUGETLB_PAGE
578 		atomic_long_t hugetlb_usage;
579 #endif
580 		struct work_struct async_put_work;
581 
582 #ifdef CONFIG_IOMMU_SUPPORT
583 		u32 pasid;
584 #endif
585 	} __randomize_layout;
586 
587 	/*
588 	 * The mm_cpumask needs to be at the end of mm_struct, because it
589 	 * is dynamically sized based on nr_cpu_ids.
590 	 */
591 	unsigned long cpu_bitmap[];
592 };
593 
594 extern struct mm_struct init_mm;
595 
596 /* Pointer magic because the dynamic array size confuses some compilers. */
597 static inline void mm_init_cpumask(struct mm_struct *mm)
598 {
599 	unsigned long cpu_bitmap = (unsigned long)mm;
600 
601 	cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
602 	cpumask_clear((struct cpumask *)cpu_bitmap);
603 }
604 
605 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
606 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
607 {
608 	return (struct cpumask *)&mm->cpu_bitmap;
609 }
610 
611 struct mmu_gather;
612 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
613 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
614 extern void tlb_finish_mmu(struct mmu_gather *tlb);
615 
616 static inline void init_tlb_flush_pending(struct mm_struct *mm)
617 {
618 	atomic_set(&mm->tlb_flush_pending, 0);
619 }
620 
621 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
622 {
623 	atomic_inc(&mm->tlb_flush_pending);
624 	/*
625 	 * The only time this value is relevant is when there are indeed pages
626 	 * to flush. And we'll only flush pages after changing them, which
627 	 * requires the PTL.
628 	 *
629 	 * So the ordering here is:
630 	 *
631 	 *	atomic_inc(&mm->tlb_flush_pending);
632 	 *	spin_lock(&ptl);
633 	 *	...
634 	 *	set_pte_at();
635 	 *	spin_unlock(&ptl);
636 	 *
637 	 *				spin_lock(&ptl)
638 	 *				mm_tlb_flush_pending();
639 	 *				....
640 	 *				spin_unlock(&ptl);
641 	 *
642 	 *	flush_tlb_range();
643 	 *	atomic_dec(&mm->tlb_flush_pending);
644 	 *
645 	 * Where the increment if constrained by the PTL unlock, it thus
646 	 * ensures that the increment is visible if the PTE modification is
647 	 * visible. After all, if there is no PTE modification, nobody cares
648 	 * about TLB flushes either.
649 	 *
650 	 * This very much relies on users (mm_tlb_flush_pending() and
651 	 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
652 	 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
653 	 * locks (PPC) the unlock of one doesn't order against the lock of
654 	 * another PTL.
655 	 *
656 	 * The decrement is ordered by the flush_tlb_range(), such that
657 	 * mm_tlb_flush_pending() will not return false unless all flushes have
658 	 * completed.
659 	 */
660 }
661 
662 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
663 {
664 	/*
665 	 * See inc_tlb_flush_pending().
666 	 *
667 	 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
668 	 * not order against TLB invalidate completion, which is what we need.
669 	 *
670 	 * Therefore we must rely on tlb_flush_*() to guarantee order.
671 	 */
672 	atomic_dec(&mm->tlb_flush_pending);
673 }
674 
675 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
676 {
677 	/*
678 	 * Must be called after having acquired the PTL; orders against that
679 	 * PTLs release and therefore ensures that if we observe the modified
680 	 * PTE we must also observe the increment from inc_tlb_flush_pending().
681 	 *
682 	 * That is, it only guarantees to return true if there is a flush
683 	 * pending for _this_ PTL.
684 	 */
685 	return atomic_read(&mm->tlb_flush_pending);
686 }
687 
688 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
689 {
690 	/*
691 	 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
692 	 * for which there is a TLB flush pending in order to guarantee
693 	 * we've seen both that PTE modification and the increment.
694 	 *
695 	 * (no requirement on actually still holding the PTL, that is irrelevant)
696 	 */
697 	return atomic_read(&mm->tlb_flush_pending) > 1;
698 }
699 
700 struct vm_fault;
701 
702 /**
703  * typedef vm_fault_t - Return type for page fault handlers.
704  *
705  * Page fault handlers return a bitmask of %VM_FAULT values.
706  */
707 typedef __bitwise unsigned int vm_fault_t;
708 
709 /**
710  * enum vm_fault_reason - Page fault handlers return a bitmask of
711  * these values to tell the core VM what happened when handling the
712  * fault. Used to decide whether a process gets delivered SIGBUS or
713  * just gets major/minor fault counters bumped up.
714  *
715  * @VM_FAULT_OOM:		Out Of Memory
716  * @VM_FAULT_SIGBUS:		Bad access
717  * @VM_FAULT_MAJOR:		Page read from storage
718  * @VM_FAULT_WRITE:		Special case for get_user_pages
719  * @VM_FAULT_HWPOISON:		Hit poisoned small page
720  * @VM_FAULT_HWPOISON_LARGE:	Hit poisoned large page. Index encoded
721  *				in upper bits
722  * @VM_FAULT_SIGSEGV:		segmentation fault
723  * @VM_FAULT_NOPAGE:		->fault installed the pte, not return page
724  * @VM_FAULT_LOCKED:		->fault locked the returned page
725  * @VM_FAULT_RETRY:		->fault blocked, must retry
726  * @VM_FAULT_FALLBACK:		huge page fault failed, fall back to small
727  * @VM_FAULT_DONE_COW:		->fault has fully handled COW
728  * @VM_FAULT_NEEDDSYNC:		->fault did not modify page tables and needs
729  *				fsync() to complete (for synchronous page faults
730  *				in DAX)
731  * @VM_FAULT_HINDEX_MASK:	mask HINDEX value
732  *
733  */
734 enum vm_fault_reason {
735 	VM_FAULT_OOM            = (__force vm_fault_t)0x000001,
736 	VM_FAULT_SIGBUS         = (__force vm_fault_t)0x000002,
737 	VM_FAULT_MAJOR          = (__force vm_fault_t)0x000004,
738 	VM_FAULT_WRITE          = (__force vm_fault_t)0x000008,
739 	VM_FAULT_HWPOISON       = (__force vm_fault_t)0x000010,
740 	VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
741 	VM_FAULT_SIGSEGV        = (__force vm_fault_t)0x000040,
742 	VM_FAULT_NOPAGE         = (__force vm_fault_t)0x000100,
743 	VM_FAULT_LOCKED         = (__force vm_fault_t)0x000200,
744 	VM_FAULT_RETRY          = (__force vm_fault_t)0x000400,
745 	VM_FAULT_FALLBACK       = (__force vm_fault_t)0x000800,
746 	VM_FAULT_DONE_COW       = (__force vm_fault_t)0x001000,
747 	VM_FAULT_NEEDDSYNC      = (__force vm_fault_t)0x002000,
748 	VM_FAULT_HINDEX_MASK    = (__force vm_fault_t)0x0f0000,
749 };
750 
751 /* Encode hstate index for a hwpoisoned large page */
752 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
753 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
754 
755 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS |	\
756 			VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON |	\
757 			VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
758 
759 #define VM_FAULT_RESULT_TRACE \
760 	{ VM_FAULT_OOM,                 "OOM" },	\
761 	{ VM_FAULT_SIGBUS,              "SIGBUS" },	\
762 	{ VM_FAULT_MAJOR,               "MAJOR" },	\
763 	{ VM_FAULT_WRITE,               "WRITE" },	\
764 	{ VM_FAULT_HWPOISON,            "HWPOISON" },	\
765 	{ VM_FAULT_HWPOISON_LARGE,      "HWPOISON_LARGE" },	\
766 	{ VM_FAULT_SIGSEGV,             "SIGSEGV" },	\
767 	{ VM_FAULT_NOPAGE,              "NOPAGE" },	\
768 	{ VM_FAULT_LOCKED,              "LOCKED" },	\
769 	{ VM_FAULT_RETRY,               "RETRY" },	\
770 	{ VM_FAULT_FALLBACK,            "FALLBACK" },	\
771 	{ VM_FAULT_DONE_COW,            "DONE_COW" },	\
772 	{ VM_FAULT_NEEDDSYNC,           "NEEDDSYNC" }
773 
774 struct vm_special_mapping {
775 	const char *name;	/* The name, e.g. "[vdso]". */
776 
777 	/*
778 	 * If .fault is not provided, this points to a
779 	 * NULL-terminated array of pages that back the special mapping.
780 	 *
781 	 * This must not be NULL unless .fault is provided.
782 	 */
783 	struct page **pages;
784 
785 	/*
786 	 * If non-NULL, then this is called to resolve page faults
787 	 * on the special mapping.  If used, .pages is not checked.
788 	 */
789 	vm_fault_t (*fault)(const struct vm_special_mapping *sm,
790 				struct vm_area_struct *vma,
791 				struct vm_fault *vmf);
792 
793 	int (*mremap)(const struct vm_special_mapping *sm,
794 		     struct vm_area_struct *new_vma);
795 };
796 
797 enum tlb_flush_reason {
798 	TLB_FLUSH_ON_TASK_SWITCH,
799 	TLB_REMOTE_SHOOTDOWN,
800 	TLB_LOCAL_SHOOTDOWN,
801 	TLB_LOCAL_MM_SHOOTDOWN,
802 	TLB_REMOTE_SEND_IPI,
803 	NR_TLB_FLUSH_REASONS,
804 };
805 
806  /*
807   * A swap entry has to fit into a "unsigned long", as the entry is hidden
808   * in the "index" field of the swapper address space.
809   */
810 typedef struct {
811 	unsigned long val;
812 } swp_entry_t;
813 
814 #endif /* _LINUX_MM_TYPES_H */
815