• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/android_kabi.h>
18 #include <linux/android_vendor.h>
19 
20 #include <asm/mmu.h>
21 
22 #ifndef AT_VECTOR_SIZE_ARCH
23 #define AT_VECTOR_SIZE_ARCH 0
24 #endif
25 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
26 
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 			 * pgdat->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 			 * @dma_addr: might require a 64-bit value even on
101 			 * 32-bit architectures.
102 			 */
103 			dma_addr_t dma_addr;
104 		};
105 		struct {	/* slab, slob and slub */
106 			union {
107 				struct list_head slab_list;
108 				struct {	/* Partial pages */
109 					struct page *next;
110 #ifdef CONFIG_64BIT
111 					int pages;	/* Nr of pages left */
112 					int pobjects;	/* Approximate count */
113 #else
114 					short int pages;
115 					short int pobjects;
116 #endif
117 				};
118 			};
119 			struct kmem_cache *slab_cache; /* not slob */
120 			/* Double-word boundary */
121 			void *freelist;		/* first free object */
122 			union {
123 				void *s_mem;	/* slab: first object */
124 				unsigned long counters;		/* SLUB */
125 				struct {			/* SLUB */
126 					unsigned inuse:16;
127 					unsigned objects:15;
128 					unsigned frozen:1;
129 				};
130 			};
131 		};
132 		struct {	/* Tail pages of compound page */
133 			unsigned long compound_head;	/* Bit zero is set */
134 
135 			/* First tail page only */
136 			unsigned char compound_dtor;
137 			unsigned char compound_order;
138 			atomic_t compound_mapcount;
139 		};
140 		struct {	/* Second tail page of compound page */
141 			unsigned long _compound_pad_1;	/* compound_head */
142 			unsigned long _compound_pad_2;
143 			/* For both global and memcg */
144 			struct list_head deferred_list;
145 		};
146 		struct {	/* Page table pages */
147 			unsigned long _pt_pad_1;	/* compound_head */
148 			pgtable_t pmd_huge_pte; /* protected by page->ptl */
149 			unsigned long _pt_pad_2;	/* mapping */
150 			union {
151 				struct mm_struct *pt_mm; /* x86 pgds only */
152 				atomic_t pt_frag_refcount; /* powerpc */
153 			};
154 #if ALLOC_SPLIT_PTLOCKS
155 			spinlock_t *ptl;
156 #else
157 			spinlock_t ptl;
158 #endif
159 		};
160 		struct {	/* ZONE_DEVICE pages */
161 			/** @pgmap: Points to the hosting device page map. */
162 			struct dev_pagemap *pgmap;
163 			void *zone_device_data;
164 			/*
165 			 * ZONE_DEVICE private pages are counted as being
166 			 * mapped so the next 3 words hold the mapping, index,
167 			 * and private fields from the source anonymous or
168 			 * page cache page while the page is migrated to device
169 			 * private memory.
170 			 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
171 			 * use the mapping, index, and private fields when
172 			 * pmem backed DAX files are mapped.
173 			 */
174 		};
175 
176 		/** @rcu_head: You can use this to free a page by RCU. */
177 		struct rcu_head rcu_head;
178 	};
179 
180 	union {		/* This union is 4 bytes in size. */
181 		/*
182 		 * If the page can be mapped to userspace, encodes the number
183 		 * of times this page is referenced by a page table.
184 		 */
185 		atomic_t _mapcount;
186 
187 		/*
188 		 * If the page is neither PageSlab nor mappable to userspace,
189 		 * the value stored here may help determine what this page
190 		 * is used for.  See page-flags.h for a list of page types
191 		 * which are currently stored here.
192 		 */
193 		unsigned int page_type;
194 
195 		unsigned int active;		/* SLAB */
196 		int units;			/* SLOB */
197 	};
198 
199 	/* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
200 	atomic_t _refcount;
201 
202 #ifdef CONFIG_MEMCG
203 	struct mem_cgroup *mem_cgroup;
204 #endif
205 
206 	/*
207 	 * On machines where all RAM is mapped into kernel address space,
208 	 * we can simply calculate the virtual address. On machines with
209 	 * highmem some memory is mapped into kernel virtual memory
210 	 * dynamically, so we need a place to store that address.
211 	 * Note that this field could be 16 bits on x86 ... ;)
212 	 *
213 	 * Architectures with slow multiplication can define
214 	 * WANT_PAGE_VIRTUAL in asm/page.h
215 	 */
216 #if defined(WANT_PAGE_VIRTUAL)
217 	void *virtual;			/* Kernel virtual address (NULL if
218 					   not kmapped, ie. highmem) */
219 #endif /* WANT_PAGE_VIRTUAL */
220 
221 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
222 	int _last_cpupid;
223 #endif
224 } _struct_page_alignment;
225 
compound_mapcount_ptr(struct page * page)226 static inline atomic_t *compound_mapcount_ptr(struct page *page)
227 {
228 	return &page[1].compound_mapcount;
229 }
230 
231 /*
232  * Used for sizing the vmemmap region on some architectures
233  */
234 #define STRUCT_PAGE_MAX_SHIFT	(order_base_2(sizeof(struct page)))
235 
236 #define PAGE_FRAG_CACHE_MAX_SIZE	__ALIGN_MASK(32768, ~PAGE_MASK)
237 #define PAGE_FRAG_CACHE_MAX_ORDER	get_order(PAGE_FRAG_CACHE_MAX_SIZE)
238 
239 #define page_private(page)		((page)->private)
240 #define set_page_private(page, v)	((page)->private = (v))
241 
242 struct page_frag_cache {
243 	void * va;
244 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
245 	__u16 offset;
246 	__u16 size;
247 #else
248 	__u32 offset;
249 #endif
250 	/* we maintain a pagecount bias, so that we dont dirty cache line
251 	 * containing page->_refcount every time we allocate a fragment.
252 	 */
253 	unsigned int		pagecnt_bias;
254 	bool pfmemalloc;
255 };
256 
257 typedef unsigned long vm_flags_t;
258 
259 /*
260  * A region containing a mapping of a non-memory backed file under NOMMU
261  * conditions.  These are held in a global tree and are pinned by the VMAs that
262  * map parts of them.
263  */
264 struct vm_region {
265 	struct rb_node	vm_rb;		/* link in global region tree */
266 	vm_flags_t	vm_flags;	/* VMA vm_flags */
267 	unsigned long	vm_start;	/* start address of region */
268 	unsigned long	vm_end;		/* region initialised to here */
269 	unsigned long	vm_top;		/* region allocated to here */
270 	unsigned long	vm_pgoff;	/* the offset in vm_file corresponding to vm_start */
271 	struct file	*vm_file;	/* the backing file or NULL */
272 
273 	int		vm_usage;	/* region usage count (access under nommu_region_sem) */
274 	bool		vm_icache_flushed : 1; /* true if the icache has been flushed for
275 						* this region */
276 };
277 
278 #ifdef CONFIG_USERFAULTFD
279 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
280 struct vm_userfaultfd_ctx {
281 	struct userfaultfd_ctx *ctx;
282 };
283 #else /* CONFIG_USERFAULTFD */
284 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
285 struct vm_userfaultfd_ctx {};
286 #endif /* CONFIG_USERFAULTFD */
287 
288 /*
289  * This struct defines a memory VMM memory area. There is one of these
290  * per VM-area/task.  A VM area is any part of the process virtual memory
291  * space that has a special rule for the page-fault handlers (ie a shared
292  * library, the executable area etc).
293  */
294 struct vm_area_struct {
295 	/* The first cache line has the info for VMA tree walking. */
296 
297 	unsigned long vm_start;		/* Our start address within vm_mm. */
298 	unsigned long vm_end;		/* The first byte after our end address
299 					   within vm_mm. */
300 
301 	/* linked list of VM areas per task, sorted by address */
302 	struct vm_area_struct *vm_next, *vm_prev;
303 
304 	struct rb_node vm_rb;
305 
306 	/*
307 	 * Largest free memory gap in bytes to the left of this VMA.
308 	 * Either between this VMA and vma->vm_prev, or between one of the
309 	 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
310 	 * get_unmapped_area find a free area of the right size.
311 	 */
312 	unsigned long rb_subtree_gap;
313 
314 	/* Second cache line starts here. */
315 
316 	struct mm_struct *vm_mm;	/* The address space we belong to. */
317 	pgprot_t vm_page_prot;		/* Access permissions of this VMA. */
318 	unsigned long vm_flags;		/* Flags, see mm.h. */
319 
320 	/*
321 	 * For areas with an address space and backing store,
322 	 * linkage into the address_space->i_mmap interval tree.
323 	 *
324 	 * For private anonymous mappings, a pointer to a null terminated string
325 	 * in the user process containing the name given to the vma, or NULL
326 	 * if unnamed.
327 	 */
328 	union {
329 		struct {
330 			struct rb_node rb;
331 			unsigned long rb_subtree_last;
332 		} shared;
333 		const char __user *anon_name;
334 	};
335 
336 	/*
337 	 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
338 	 * list, after a COW of one of the file pages.	A MAP_SHARED vma
339 	 * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
340 	 * or brk vma (with NULL file) can only be in an anon_vma list.
341 	 */
342 	struct list_head anon_vma_chain; /* Serialized by mmap_sem &
343 					  * page_table_lock */
344 	struct anon_vma *anon_vma;	/* Serialized by page_table_lock */
345 
346 	/* Function pointers to deal with this struct. */
347 	const struct vm_operations_struct *vm_ops;
348 
349 	/* Information about our backing store: */
350 	unsigned long vm_pgoff;		/* Offset (within vm_file) in PAGE_SIZE
351 					   units */
352 	struct file * vm_file;		/* File we map to (can be NULL). */
353 	void * vm_private_data;		/* was vm_pte (shared mem) */
354 
355 #ifdef CONFIG_SWAP
356 	atomic_long_t swap_readahead_info;
357 #endif
358 #ifndef CONFIG_MMU
359 	struct vm_region *vm_region;	/* NOMMU mapping region */
360 #endif
361 #ifdef CONFIG_NUMA
362 	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
363 #endif
364 	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
365 
366 	ANDROID_KABI_RESERVE(1);
367 	ANDROID_KABI_RESERVE(2);
368 	ANDROID_KABI_RESERVE(3);
369 	ANDROID_KABI_RESERVE(4);
370 	ANDROID_VENDOR_DATA(1);
371 } __randomize_layout;
372 
373 struct core_thread {
374 	struct task_struct *task;
375 	struct core_thread *next;
376 };
377 
378 struct core_state {
379 	atomic_t nr_threads;
380 	struct core_thread dumper;
381 	struct completion startup;
382 };
383 
384 struct kioctx_table;
385 struct mm_struct {
386 	struct {
387 		struct vm_area_struct *mmap;		/* list of VMAs */
388 		struct rb_root mm_rb;
389 		u64 vmacache_seqnum;                   /* per-thread vmacache */
390 #ifdef CONFIG_MMU
391 		unsigned long (*get_unmapped_area) (struct file *filp,
392 				unsigned long addr, unsigned long len,
393 				unsigned long pgoff, unsigned long flags);
394 #endif
395 		unsigned long mmap_base;	/* base of mmap area */
396 		unsigned long mmap_legacy_base;	/* base of mmap area in bottom-up allocations */
397 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
398 		/* Base adresses for compatible mmap() */
399 		unsigned long mmap_compat_base;
400 		unsigned long mmap_compat_legacy_base;
401 #endif
402 		unsigned long task_size;	/* size of task vm space */
403 		unsigned long highest_vm_end;	/* highest vma end address */
404 		pgd_t * pgd;
405 
406 #ifdef CONFIG_MEMBARRIER
407 		/**
408 		 * @membarrier_state: Flags controlling membarrier behavior.
409 		 *
410 		 * This field is close to @pgd to hopefully fit in the same
411 		 * cache-line, which needs to be touched by switch_mm().
412 		 */
413 		atomic_t membarrier_state;
414 #endif
415 
416 		/**
417 		 * @mm_users: The number of users including userspace.
418 		 *
419 		 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
420 		 * drops to 0 (i.e. when the task exits and there are no other
421 		 * temporary reference holders), we also release a reference on
422 		 * @mm_count (which may then free the &struct mm_struct if
423 		 * @mm_count also drops to 0).
424 		 */
425 		atomic_t mm_users;
426 
427 		/**
428 		 * @mm_count: The number of references to &struct mm_struct
429 		 * (@mm_users count as 1).
430 		 *
431 		 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
432 		 * &struct mm_struct is freed.
433 		 */
434 		atomic_t mm_count;
435 
436 #ifdef CONFIG_MMU
437 		atomic_long_t pgtables_bytes;	/* PTE page table pages */
438 #endif
439 		int map_count;			/* number of VMAs */
440 
441 		spinlock_t page_table_lock; /* Protects page tables and some
442 					     * counters
443 					     */
444 		struct rw_semaphore mmap_sem;
445 
446 		struct list_head mmlist; /* List of maybe swapped mm's.	These
447 					  * are globally strung together off
448 					  * init_mm.mmlist, and are protected
449 					  * by mmlist_lock
450 					  */
451 
452 
453 		unsigned long hiwater_rss; /* High-watermark of RSS usage */
454 		unsigned long hiwater_vm;  /* High-water virtual memory usage */
455 
456 		unsigned long total_vm;	   /* Total pages mapped */
457 		unsigned long locked_vm;   /* Pages that have PG_mlocked set */
458 		atomic64_t    pinned_vm;   /* Refcount permanently increased */
459 		unsigned long data_vm;	   /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
460 		unsigned long exec_vm;	   /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
461 		unsigned long stack_vm;	   /* VM_STACK */
462 		unsigned long def_flags;
463 
464 		spinlock_t arg_lock; /* protect the below fields */
465 		unsigned long start_code, end_code, start_data, end_data;
466 		unsigned long start_brk, brk, start_stack;
467 		unsigned long arg_start, arg_end, env_start, env_end;
468 
469 		unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
470 
471 		/*
472 		 * Special counters, in some configurations protected by the
473 		 * page_table_lock, in other configurations by being atomic.
474 		 */
475 		struct mm_rss_stat rss_stat;
476 
477 		struct linux_binfmt *binfmt;
478 
479 		/* Architecture-specific MM context */
480 		mm_context_t context;
481 
482 		unsigned long flags; /* Must use atomic bitops to access */
483 
484 		struct core_state *core_state; /* coredumping support */
485 
486 #ifdef CONFIG_AIO
487 		spinlock_t			ioctx_lock;
488 		struct kioctx_table __rcu	*ioctx_table;
489 #endif
490 #ifdef CONFIG_MEMCG
491 		/*
492 		 * "owner" points to a task that is regarded as the canonical
493 		 * user/owner of this mm. All of the following must be true in
494 		 * order for it to be changed:
495 		 *
496 		 * current == mm->owner
497 		 * current->mm != mm
498 		 * new_owner->mm == mm
499 		 * new_owner->alloc_lock is held
500 		 */
501 		struct task_struct __rcu *owner;
502 #endif
503 		struct user_namespace *user_ns;
504 
505 		/* store ref to file /proc/<pid>/exe symlink points to */
506 		struct file __rcu *exe_file;
507 #ifdef CONFIG_MMU_NOTIFIER
508 		struct mmu_notifier_mm *mmu_notifier_mm;
509 #endif
510 #if (defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_GKI_OPT_FEATURES)) && \
511     !USE_SPLIT_PMD_PTLOCKS
512 		pgtable_t pmd_huge_pte; /* protected by page_table_lock */
513 #endif
514 #ifdef CONFIG_NUMA_BALANCING
515 		/*
516 		 * numa_next_scan is the next time that the PTEs will be marked
517 		 * pte_numa. NUMA hinting faults will gather statistics and
518 		 * migrate pages to new nodes if necessary.
519 		 */
520 		unsigned long numa_next_scan;
521 
522 		/* Restart point for scanning and setting pte_numa */
523 		unsigned long numa_scan_offset;
524 
525 		/* numa_scan_seq prevents two threads setting pte_numa */
526 		int numa_scan_seq;
527 #endif
528 		/*
529 		 * An operation with batched TLB flushing is going on. Anything
530 		 * that can move process memory needs to flush the TLB when
531 		 * moving a PROT_NONE or PROT_NUMA mapped page.
532 		 */
533 		atomic_t tlb_flush_pending;
534 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
535 		/* See flush_tlb_batched_pending() */
536 		bool tlb_flush_batched;
537 #endif
538 		struct uprobes_state uprobes_state;
539 #ifdef CONFIG_HUGETLB_PAGE
540 		atomic_long_t hugetlb_usage;
541 #endif
542 		struct work_struct async_put_work;
543 		ANDROID_VENDOR_DATA(1);
544 	} __randomize_layout;
545 
546 	/*
547 	 * The mm_cpumask needs to be at the end of mm_struct, because it
548 	 * is dynamically sized based on nr_cpu_ids.
549 	 */
550 	unsigned long cpu_bitmap[];
551 };
552 
553 extern struct mm_struct init_mm;
554 
555 /* Pointer magic because the dynamic array size confuses some compilers. */
mm_init_cpumask(struct mm_struct * mm)556 static inline void mm_init_cpumask(struct mm_struct *mm)
557 {
558 	unsigned long cpu_bitmap = (unsigned long)mm;
559 
560 	cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
561 	cpumask_clear((struct cpumask *)cpu_bitmap);
562 }
563 
564 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
mm_cpumask(struct mm_struct * mm)565 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
566 {
567 	return (struct cpumask *)&mm->cpu_bitmap;
568 }
569 
570 struct mmu_gather;
571 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
572 				unsigned long start, unsigned long end);
573 extern void tlb_finish_mmu(struct mmu_gather *tlb,
574 				unsigned long start, unsigned long end);
575 
init_tlb_flush_pending(struct mm_struct * mm)576 static inline void init_tlb_flush_pending(struct mm_struct *mm)
577 {
578 	atomic_set(&mm->tlb_flush_pending, 0);
579 }
580 
inc_tlb_flush_pending(struct mm_struct * mm)581 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
582 {
583 	atomic_inc(&mm->tlb_flush_pending);
584 	/*
585 	 * The only time this value is relevant is when there are indeed pages
586 	 * to flush. And we'll only flush pages after changing them, which
587 	 * requires the PTL.
588 	 *
589 	 * So the ordering here is:
590 	 *
591 	 *	atomic_inc(&mm->tlb_flush_pending);
592 	 *	spin_lock(&ptl);
593 	 *	...
594 	 *	set_pte_at();
595 	 *	spin_unlock(&ptl);
596 	 *
597 	 *				spin_lock(&ptl)
598 	 *				mm_tlb_flush_pending();
599 	 *				....
600 	 *				spin_unlock(&ptl);
601 	 *
602 	 *	flush_tlb_range();
603 	 *	atomic_dec(&mm->tlb_flush_pending);
604 	 *
605 	 * Where the increment if constrained by the PTL unlock, it thus
606 	 * ensures that the increment is visible if the PTE modification is
607 	 * visible. After all, if there is no PTE modification, nobody cares
608 	 * about TLB flushes either.
609 	 *
610 	 * This very much relies on users (mm_tlb_flush_pending() and
611 	 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
612 	 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
613 	 * locks (PPC) the unlock of one doesn't order against the lock of
614 	 * another PTL.
615 	 *
616 	 * The decrement is ordered by the flush_tlb_range(), such that
617 	 * mm_tlb_flush_pending() will not return false unless all flushes have
618 	 * completed.
619 	 */
620 }
621 
dec_tlb_flush_pending(struct mm_struct * mm)622 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
623 {
624 	/*
625 	 * See inc_tlb_flush_pending().
626 	 *
627 	 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
628 	 * not order against TLB invalidate completion, which is what we need.
629 	 *
630 	 * Therefore we must rely on tlb_flush_*() to guarantee order.
631 	 */
632 	atomic_dec(&mm->tlb_flush_pending);
633 }
634 
mm_tlb_flush_pending(struct mm_struct * mm)635 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
636 {
637 	/*
638 	 * Must be called after having acquired the PTL; orders against that
639 	 * PTLs release and therefore ensures that if we observe the modified
640 	 * PTE we must also observe the increment from inc_tlb_flush_pending().
641 	 *
642 	 * That is, it only guarantees to return true if there is a flush
643 	 * pending for _this_ PTL.
644 	 */
645 	return atomic_read(&mm->tlb_flush_pending);
646 }
647 
mm_tlb_flush_nested(struct mm_struct * mm)648 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
649 {
650 	/*
651 	 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
652 	 * for which there is a TLB flush pending in order to guarantee
653 	 * we've seen both that PTE modification and the increment.
654 	 *
655 	 * (no requirement on actually still holding the PTL, that is irrelevant)
656 	 */
657 	return atomic_read(&mm->tlb_flush_pending) > 1;
658 }
659 
660 struct vm_fault;
661 
662 /**
663  * typedef vm_fault_t - Return type for page fault handlers.
664  *
665  * Page fault handlers return a bitmask of %VM_FAULT values.
666  */
667 typedef __bitwise unsigned int vm_fault_t;
668 
669 /**
670  * enum vm_fault_reason - Page fault handlers return a bitmask of
671  * these values to tell the core VM what happened when handling the
672  * fault. Used to decide whether a process gets delivered SIGBUS or
673  * just gets major/minor fault counters bumped up.
674  *
675  * @VM_FAULT_OOM:		Out Of Memory
676  * @VM_FAULT_SIGBUS:		Bad access
677  * @VM_FAULT_MAJOR:		Page read from storage
678  * @VM_FAULT_WRITE:		Special case for get_user_pages
679  * @VM_FAULT_HWPOISON:		Hit poisoned small page
680  * @VM_FAULT_HWPOISON_LARGE:	Hit poisoned large page. Index encoded
681  *				in upper bits
682  * @VM_FAULT_SIGSEGV:		segmentation fault
683  * @VM_FAULT_NOPAGE:		->fault installed the pte, not return page
684  * @VM_FAULT_LOCKED:		->fault locked the returned page
685  * @VM_FAULT_RETRY:		->fault blocked, must retry
686  * @VM_FAULT_FALLBACK:		huge page fault failed, fall back to small
687  * @VM_FAULT_DONE_COW:		->fault has fully handled COW
688  * @VM_FAULT_NEEDDSYNC:		->fault did not modify page tables and needs
689  *				fsync() to complete (for synchronous page faults
690  *				in DAX)
691  * @VM_FAULT_HINDEX_MASK:	mask HINDEX value
692  *
693  */
694 enum vm_fault_reason {
695 	VM_FAULT_OOM            = (__force vm_fault_t)0x000001,
696 	VM_FAULT_SIGBUS         = (__force vm_fault_t)0x000002,
697 	VM_FAULT_MAJOR          = (__force vm_fault_t)0x000004,
698 	VM_FAULT_WRITE          = (__force vm_fault_t)0x000008,
699 	VM_FAULT_HWPOISON       = (__force vm_fault_t)0x000010,
700 	VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
701 	VM_FAULT_SIGSEGV        = (__force vm_fault_t)0x000040,
702 	VM_FAULT_NOPAGE         = (__force vm_fault_t)0x000100,
703 	VM_FAULT_LOCKED         = (__force vm_fault_t)0x000200,
704 	VM_FAULT_RETRY          = (__force vm_fault_t)0x000400,
705 	VM_FAULT_FALLBACK       = (__force vm_fault_t)0x000800,
706 	VM_FAULT_DONE_COW       = (__force vm_fault_t)0x001000,
707 	VM_FAULT_NEEDDSYNC      = (__force vm_fault_t)0x002000,
708 	VM_FAULT_HINDEX_MASK    = (__force vm_fault_t)0x0f0000,
709 };
710 
711 /* Encode hstate index for a hwpoisoned large page */
712 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
713 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
714 
715 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS |	\
716 			VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON |	\
717 			VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
718 
719 #define VM_FAULT_RESULT_TRACE \
720 	{ VM_FAULT_OOM,                 "OOM" },	\
721 	{ VM_FAULT_SIGBUS,              "SIGBUS" },	\
722 	{ VM_FAULT_MAJOR,               "MAJOR" },	\
723 	{ VM_FAULT_WRITE,               "WRITE" },	\
724 	{ VM_FAULT_HWPOISON,            "HWPOISON" },	\
725 	{ VM_FAULT_HWPOISON_LARGE,      "HWPOISON_LARGE" },	\
726 	{ VM_FAULT_SIGSEGV,             "SIGSEGV" },	\
727 	{ VM_FAULT_NOPAGE,              "NOPAGE" },	\
728 	{ VM_FAULT_LOCKED,              "LOCKED" },	\
729 	{ VM_FAULT_RETRY,               "RETRY" },	\
730 	{ VM_FAULT_FALLBACK,            "FALLBACK" },	\
731 	{ VM_FAULT_DONE_COW,            "DONE_COW" },	\
732 	{ VM_FAULT_NEEDDSYNC,           "NEEDDSYNC" }
733 
734 struct vm_special_mapping {
735 	const char *name;	/* The name, e.g. "[vdso]". */
736 
737 	/*
738 	 * If .fault is not provided, this points to a
739 	 * NULL-terminated array of pages that back the special mapping.
740 	 *
741 	 * This must not be NULL unless .fault is provided.
742 	 */
743 	struct page **pages;
744 
745 	/*
746 	 * If non-NULL, then this is called to resolve page faults
747 	 * on the special mapping.  If used, .pages is not checked.
748 	 */
749 	vm_fault_t (*fault)(const struct vm_special_mapping *sm,
750 				struct vm_area_struct *vma,
751 				struct vm_fault *vmf);
752 
753 	int (*mremap)(const struct vm_special_mapping *sm,
754 		     struct vm_area_struct *new_vma);
755 };
756 
757 enum tlb_flush_reason {
758 	TLB_FLUSH_ON_TASK_SWITCH,
759 	TLB_REMOTE_SHOOTDOWN,
760 	TLB_LOCAL_SHOOTDOWN,
761 	TLB_LOCAL_MM_SHOOTDOWN,
762 	TLB_REMOTE_SEND_IPI,
763 	NR_TLB_FLUSH_REASONS,
764 };
765 
766  /*
767   * A swap entry has to fit into a "unsigned long", as the entry is hidden
768   * in the "index" field of the swapper address space.
769   */
770 typedef struct {
771 	unsigned long val;
772 } swp_entry_t;
773 
774 /* Return the name for an anonymous mapping or NULL for a file-backed mapping */
vma_get_anon_name(struct vm_area_struct * vma)775 static inline const char __user *vma_get_anon_name(struct vm_area_struct *vma)
776 {
777 	if (vma->vm_file)
778 		return NULL;
779 
780 	return vma->anon_name;
781 }
782 
783 #endif /* _LINUX_MM_TYPES_H */
784