• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 #ifndef _LINUX_MM_TYPES_H
2 #define _LINUX_MM_TYPES_H
3 
4 #include <linux/auxvec.h>
5 #include <linux/types.h>
6 #include <linux/threads.h>
7 #include <linux/list.h>
8 #include <linux/spinlock.h>
9 #include <linux/rbtree.h>
10 #include <linux/rwsem.h>
11 #include <linux/completion.h>
12 #include <linux/cpumask.h>
13 #include <linux/page-debug-flags.h>
14 #include <linux/uprobes.h>
15 #include <linux/page-flags-layout.h>
16 #include <linux/workqueue.h>
17 #include <asm/page.h>
18 #include <asm/mmu.h>
19 
20 #ifndef AT_VECTOR_SIZE_ARCH
21 #define AT_VECTOR_SIZE_ARCH 0
22 #endif
23 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
24 
25 struct address_space;
26 
27 #define USE_SPLIT_PTE_PTLOCKS	(NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
28 #define USE_SPLIT_PMD_PTLOCKS	(USE_SPLIT_PTE_PTLOCKS && \
29 		IS_ENABLED(CONFIG_ARCH_ENABLE_SPLIT_PMD_PTLOCK))
30 #define ALLOC_SPLIT_PTLOCKS	(SPINLOCK_SIZE > BITS_PER_LONG/8)
31 
32 /*
33  * Each physical page in the system has a struct page associated with
34  * it to keep track of whatever it is we are using the page for at the
35  * moment. Note that we have no way to track which tasks are using
36  * a page, though if it is a pagecache page, rmap structures can tell us
37  * who is mapping it.
38  *
39  * The objects in struct page are organized in double word blocks in
40  * order to allows us to use atomic double word operations on portions
41  * of struct page. That is currently only used by slub but the arrangement
42  * allows the use of atomic double word operations on the flags/mapping
43  * and lru list pointers also.
44  */
45 struct page {
46 	/* First double word block */
47 	unsigned long flags;		/* Atomic flags, some possibly
48 					 * updated asynchronously */
49 	union {
50 		struct address_space *mapping;	/* If low bit clear, points to
51 						 * inode address_space, or NULL.
52 						 * If page mapped as anonymous
53 						 * memory, low bit is set, and
54 						 * it points to anon_vma object:
55 						 * see PAGE_MAPPING_ANON below.
56 						 */
57 		void *s_mem;			/* slab first object */
58 	};
59 
60 	/* Second double word */
61 	struct {
62 		union {
63 			pgoff_t index;		/* Our offset within mapping. */
64 			void *freelist;		/* sl[aou]b first free object */
65 			bool pfmemalloc;	/* If set by the page allocator,
66 						 * ALLOC_NO_WATERMARKS was set
67 						 * and the low watermark was not
68 						 * met implying that the system
69 						 * is under some pressure. The
70 						 * caller should try ensure
71 						 * this page is only used to
72 						 * free other pages.
73 						 */
74 		};
75 
76 		union {
77 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
78 	defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
79 			/* Used for cmpxchg_double in slub */
80 			unsigned long counters;
81 #else
82 			/*
83 			 * Keep _count separate from slub cmpxchg_double data.
84 			 * As the rest of the double word is protected by
85 			 * slab_lock but _count is not.
86 			 */
87 			unsigned counters;
88 #endif
89 
90 			struct {
91 
92 				union {
93 					/*
94 					 * Count of ptes mapped in
95 					 * mms, to show when page is
96 					 * mapped & limit reverse map
97 					 * searches.
98 					 *
99 					 * Used also for tail pages
100 					 * refcounting instead of
101 					 * _count. Tail pages cannot
102 					 * be mapped and keeping the
103 					 * tail page _count zero at
104 					 * all times guarantees
105 					 * get_page_unless_zero() will
106 					 * never succeed on tail
107 					 * pages.
108 					 */
109 					atomic_t _mapcount;
110 
111 					struct { /* SLUB */
112 						unsigned inuse:16;
113 						unsigned objects:15;
114 						unsigned frozen:1;
115 					};
116 					int units;	/* SLOB */
117 				};
118 				atomic_t _count;		/* Usage count, see below. */
119 			};
120 			unsigned int active;	/* SLAB */
121 		};
122 	};
123 
124 	/* Third double word block */
125 	union {
126 		struct list_head lru;	/* Pageout list, eg. active_list
127 					 * protected by zone->lru_lock !
128 					 * Can be used as a generic list
129 					 * by the page owner.
130 					 */
131 		struct {		/* slub per cpu partial pages */
132 			struct page *next;	/* Next partial slab */
133 #ifdef CONFIG_64BIT
134 			int pages;	/* Nr of partial slabs left */
135 			int pobjects;	/* Approximate # of objects */
136 #else
137 			short int pages;
138 			short int pobjects;
139 #endif
140 		};
141 
142 		struct slab *slab_page; /* slab fields */
143 		struct rcu_head rcu_head;	/* Used by SLAB
144 						 * when destroying via RCU
145 						 */
146 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
147 		pgtable_t pmd_huge_pte; /* protected by page->ptl */
148 #endif
149 	};
150 
151 	/* Remainder is not double word aligned */
152 	union {
153 		unsigned long private;		/* Mapping-private opaque data:
154 					 	 * usually used for buffer_heads
155 						 * if PagePrivate set; used for
156 						 * swp_entry_t if PageSwapCache;
157 						 * indicates order in the buddy
158 						 * system if PG_buddy is set.
159 						 */
160 #if USE_SPLIT_PTE_PTLOCKS
161 #if ALLOC_SPLIT_PTLOCKS
162 		spinlock_t *ptl;
163 #else
164 		spinlock_t ptl;
165 #endif
166 #endif
167 		struct kmem_cache *slab_cache;	/* SL[AU]B: Pointer to slab */
168 		struct page *first_page;	/* Compound tail pages */
169 	};
170 
171 	/*
172 	 * On machines where all RAM is mapped into kernel address space,
173 	 * we can simply calculate the virtual address. On machines with
174 	 * highmem some memory is mapped into kernel virtual memory
175 	 * dynamically, so we need a place to store that address.
176 	 * Note that this field could be 16 bits on x86 ... ;)
177 	 *
178 	 * Architectures with slow multiplication can define
179 	 * WANT_PAGE_VIRTUAL in asm/page.h
180 	 */
181 #if defined(WANT_PAGE_VIRTUAL)
182 	void *virtual;			/* Kernel virtual address (NULL if
183 					   not kmapped, ie. highmem) */
184 #endif /* WANT_PAGE_VIRTUAL */
185 #ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS
186 	unsigned long debug_flags;	/* Use atomic bitops on this */
187 #endif
188 
189 #ifdef CONFIG_KMEMCHECK
190 	/*
191 	 * kmemcheck wants to track the status of each byte in a page; this
192 	 * is a pointer to such a status block. NULL if not tracked.
193 	 */
194 	void *shadow;
195 #endif
196 
197 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
198 	int _last_cpupid;
199 #endif
200 }
201 /*
202  * The struct page can be forced to be double word aligned so that atomic ops
203  * on double words work. The SLUB allocator can make use of such a feature.
204  */
205 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
206 	__aligned(2 * sizeof(unsigned long))
207 #endif
208 ;
209 
210 struct page_frag {
211 	struct page *page;
212 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
213 	__u32 offset;
214 	__u32 size;
215 #else
216 	__u16 offset;
217 	__u16 size;
218 #endif
219 };
220 
221 typedef unsigned long __nocast vm_flags_t;
222 
223 /*
224  * A region containing a mapping of a non-memory backed file under NOMMU
225  * conditions.  These are held in a global tree and are pinned by the VMAs that
226  * map parts of them.
227  */
228 struct vm_region {
229 	struct rb_node	vm_rb;		/* link in global region tree */
230 	vm_flags_t	vm_flags;	/* VMA vm_flags */
231 	unsigned long	vm_start;	/* start address of region */
232 	unsigned long	vm_end;		/* region initialised to here */
233 	unsigned long	vm_top;		/* region allocated to here */
234 	unsigned long	vm_pgoff;	/* the offset in vm_file corresponding to vm_start */
235 	struct file	*vm_file;	/* the backing file or NULL */
236 
237 	int		vm_usage;	/* region usage count (access under nommu_region_sem) */
238 	bool		vm_icache_flushed : 1; /* true if the icache has been flushed for
239 						* this region */
240 };
241 
242 /*
243  * This struct defines a memory VMM memory area. There is one of these
244  * per VM-area/task.  A VM area is any part of the process virtual memory
245  * space that has a special rule for the page-fault handlers (ie a shared
246  * library, the executable area etc).
247  */
248 struct vm_area_struct {
249 	/* The first cache line has the info for VMA tree walking. */
250 
251 	unsigned long vm_start;		/* Our start address within vm_mm. */
252 	unsigned long vm_end;		/* The first byte after our end address
253 					   within vm_mm. */
254 
255 	/* linked list of VM areas per task, sorted by address */
256 	struct vm_area_struct *vm_next, *vm_prev;
257 
258 	struct rb_node vm_rb;
259 
260 	/*
261 	 * Largest free memory gap in bytes to the left of this VMA.
262 	 * Either between this VMA and vma->vm_prev, or between one of the
263 	 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
264 	 * get_unmapped_area find a free area of the right size.
265 	 */
266 	unsigned long rb_subtree_gap;
267 
268 	/* Second cache line starts here. */
269 
270 	struct mm_struct *vm_mm;	/* The address space we belong to. */
271 	pgprot_t vm_page_prot;		/* Access permissions of this VMA. */
272 	unsigned long vm_flags;		/* Flags, see mm.h. */
273 
274 	/*
275 	 * For areas with an address space and backing store,
276 	 * linkage into the address_space->i_mmap interval tree, or
277 	 * linkage of vma in the address_space->i_mmap_nonlinear list.
278 	 *
279 	 * For private anonymous mappings, a pointer to a null terminated string
280 	 * in the user process containing the name given to the vma, or NULL
281 	 * if unnamed.
282 	 */
283 	union {
284 		struct {
285 			struct rb_node rb;
286 			unsigned long rb_subtree_last;
287 		} linear;
288 		struct list_head nonlinear;
289 		const char __user *anon_name;
290 	} shared;
291 
292 	/*
293 	 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
294 	 * list, after a COW of one of the file pages.	A MAP_SHARED vma
295 	 * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
296 	 * or brk vma (with NULL file) can only be in an anon_vma list.
297 	 */
298 	struct list_head anon_vma_chain; /* Serialized by mmap_sem &
299 					  * page_table_lock */
300 	struct anon_vma *anon_vma;	/* Serialized by page_table_lock */
301 
302 	/* Function pointers to deal with this struct. */
303 	const struct vm_operations_struct *vm_ops;
304 
305 	/* Information about our backing store: */
306 	unsigned long vm_pgoff;		/* Offset (within vm_file) in PAGE_SIZE
307 					   units, *not* PAGE_CACHE_SIZE */
308 	struct file * vm_file;		/* File we map to (can be NULL). */
309 	void * vm_private_data;		/* was vm_pte (shared mem) */
310 
311 #ifndef CONFIG_MMU
312 	struct vm_region *vm_region;	/* NOMMU mapping region */
313 #endif
314 #ifdef CONFIG_NUMA
315 	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
316 #endif
317 };
318 
319 struct core_thread {
320 	struct task_struct *task;
321 	struct core_thread *next;
322 };
323 
324 struct core_state {
325 	atomic_t nr_threads;
326 	struct core_thread dumper;
327 	struct completion startup;
328 };
329 
330 enum {
331 	MM_FILEPAGES,
332 	MM_ANONPAGES,
333 	MM_SWAPENTS,
334 	NR_MM_COUNTERS
335 };
336 
337 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
338 #define SPLIT_RSS_COUNTING
339 /* per-thread cached information, */
340 struct task_rss_stat {
341 	int events;	/* for synchronization threshold */
342 	int count[NR_MM_COUNTERS];
343 };
344 #endif /* USE_SPLIT_PTE_PTLOCKS */
345 
346 struct mm_rss_stat {
347 	atomic_long_t count[NR_MM_COUNTERS];
348 };
349 
350 struct kioctx_table;
351 struct mm_struct {
352 	struct vm_area_struct *mmap;		/* list of VMAs */
353 	struct rb_root mm_rb;
354 	u32 vmacache_seqnum;                   /* per-thread vmacache */
355 #ifdef CONFIG_MMU
356 	unsigned long (*get_unmapped_area) (struct file *filp,
357 				unsigned long addr, unsigned long len,
358 				unsigned long pgoff, unsigned long flags);
359 #endif
360 	unsigned long mmap_base;		/* base of mmap area */
361 	unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
362 	unsigned long task_size;		/* size of task vm space */
363 	unsigned long highest_vm_end;		/* highest vma end address */
364 	pgd_t * pgd;
365 	atomic_t mm_users;			/* How many users with user space? */
366 	atomic_t mm_count;			/* How many references to "struct mm_struct" (users count as 1) */
367 	atomic_long_t nr_ptes;			/* Page table pages */
368 	int map_count;				/* number of VMAs */
369 
370 	spinlock_t page_table_lock;		/* Protects page tables and some counters */
371 	struct rw_semaphore mmap_sem;
372 
373 	struct list_head mmlist;		/* List of maybe swapped mm's.	These are globally strung
374 						 * together off init_mm.mmlist, and are protected
375 						 * by mmlist_lock
376 						 */
377 
378 
379 	unsigned long hiwater_rss;	/* High-watermark of RSS usage */
380 	unsigned long hiwater_vm;	/* High-water virtual memory usage */
381 
382 	unsigned long total_vm;		/* Total pages mapped */
383 	unsigned long locked_vm;	/* Pages that have PG_mlocked set */
384 	unsigned long pinned_vm;	/* Refcount permanently increased */
385 	unsigned long shared_vm;	/* Shared pages (files) */
386 	unsigned long exec_vm;		/* VM_EXEC & ~VM_WRITE */
387 	unsigned long stack_vm;		/* VM_GROWSUP/DOWN */
388 	unsigned long def_flags;
389 	unsigned long start_code, end_code, start_data, end_data;
390 	unsigned long start_brk, brk, start_stack;
391 	unsigned long arg_start, arg_end, env_start, env_end;
392 
393 	unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
394 
395 	/*
396 	 * Special counters, in some configurations protected by the
397 	 * page_table_lock, in other configurations by being atomic.
398 	 */
399 	struct mm_rss_stat rss_stat;
400 
401 	struct linux_binfmt *binfmt;
402 
403 	cpumask_var_t cpu_vm_mask_var;
404 
405 	/* Architecture-specific MM context */
406 	mm_context_t context;
407 
408 	unsigned long flags; /* Must use atomic bitops to access the bits */
409 
410 	struct core_state *core_state; /* coredumping support */
411 #ifdef CONFIG_AIO
412 	spinlock_t			ioctx_lock;
413 	struct kioctx_table __rcu	*ioctx_table;
414 #endif
415 #ifdef CONFIG_MEMCG
416 	/*
417 	 * "owner" points to a task that is regarded as the canonical
418 	 * user/owner of this mm. All of the following must be true in
419 	 * order for it to be changed:
420 	 *
421 	 * current == mm->owner
422 	 * current->mm != mm
423 	 * new_owner->mm == mm
424 	 * new_owner->alloc_lock is held
425 	 */
426 	struct task_struct __rcu *owner;
427 #endif
428 
429 	/* store ref to file /proc/<pid>/exe symlink points to */
430 	struct file *exe_file;
431 #ifdef CONFIG_MMU_NOTIFIER
432 	struct mmu_notifier_mm *mmu_notifier_mm;
433 #endif
434 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
435 	pgtable_t pmd_huge_pte; /* protected by page_table_lock */
436 #endif
437 #ifdef CONFIG_CPUMASK_OFFSTACK
438 	struct cpumask cpumask_allocation;
439 #endif
440 #ifdef CONFIG_NUMA_BALANCING
441 	/*
442 	 * numa_next_scan is the next time that the PTEs will be marked
443 	 * pte_numa. NUMA hinting faults will gather statistics and migrate
444 	 * pages to new nodes if necessary.
445 	 */
446 	unsigned long numa_next_scan;
447 
448 	/* Restart point for scanning and setting pte_numa */
449 	unsigned long numa_scan_offset;
450 
451 	/* numa_scan_seq prevents two threads setting pte_numa */
452 	int numa_scan_seq;
453 #endif
454 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
455 	/*
456 	 * An operation with batched TLB flushing is going on. Anything that
457 	 * can move process memory needs to flush the TLB when moving a
458 	 * PROT_NONE or PROT_NUMA mapped page.
459 	 */
460 	bool tlb_flush_pending;
461 #endif
462 	struct uprobes_state uprobes_state;
463 	struct work_struct async_put_work;
464 };
465 
mm_init_cpumask(struct mm_struct * mm)466 static inline void mm_init_cpumask(struct mm_struct *mm)
467 {
468 #ifdef CONFIG_CPUMASK_OFFSTACK
469 	mm->cpu_vm_mask_var = &mm->cpumask_allocation;
470 #endif
471 	cpumask_clear(mm->cpu_vm_mask_var);
472 }
473 
474 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
mm_cpumask(struct mm_struct * mm)475 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
476 {
477 	return mm->cpu_vm_mask_var;
478 }
479 
480 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
481 /*
482  * Memory barriers to keep this state in sync are graciously provided by
483  * the page table locks, outside of which no page table modifications happen.
484  * The barriers below prevent the compiler from re-ordering the instructions
485  * around the memory barriers that are already present in the code.
486  */
mm_tlb_flush_pending(struct mm_struct * mm)487 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
488 {
489 	barrier();
490 	return mm->tlb_flush_pending;
491 }
set_tlb_flush_pending(struct mm_struct * mm)492 static inline void set_tlb_flush_pending(struct mm_struct *mm)
493 {
494 	mm->tlb_flush_pending = true;
495 
496 	/*
497 	 * Guarantee that the tlb_flush_pending store does not leak into the
498 	 * critical section updating the page tables
499 	 */
500 	smp_mb__before_spinlock();
501 }
502 /* Clearing is done after a TLB flush, which also provides a barrier. */
clear_tlb_flush_pending(struct mm_struct * mm)503 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
504 {
505 	barrier();
506 	mm->tlb_flush_pending = false;
507 }
508 #else
mm_tlb_flush_pending(struct mm_struct * mm)509 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
510 {
511 	return false;
512 }
set_tlb_flush_pending(struct mm_struct * mm)513 static inline void set_tlb_flush_pending(struct mm_struct *mm)
514 {
515 }
clear_tlb_flush_pending(struct mm_struct * mm)516 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
517 {
518 }
519 #endif
520 
521 struct vm_special_mapping
522 {
523 	const char *name;
524 	struct page **pages;
525 };
526 
527 enum tlb_flush_reason {
528 	TLB_FLUSH_ON_TASK_SWITCH,
529 	TLB_REMOTE_SHOOTDOWN,
530 	TLB_LOCAL_SHOOTDOWN,
531 	TLB_LOCAL_MM_SHOOTDOWN,
532 	NR_TLB_FLUSH_REASONS,
533 };
534 
535 /* Return the name for an anonymous mapping or NULL for a file-backed mapping */
vma_get_anon_name(struct vm_area_struct * vma)536 static inline const char __user *vma_get_anon_name(struct vm_area_struct *vma)
537 {
538 	if (vma->vm_file)
539 		return NULL;
540 
541 	return vma->shared.anon_name;
542 }
543 
544 #endif /* _LINUX_MM_TYPES_H */
545