xref: /linux-6.15/include/linux/mm_types.h (revision f7be3455)
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/prio_tree.h>
10 #include <linux/rbtree.h>
11 #include <linux/rwsem.h>
12 #include <linux/completion.h>
13 #include <linux/cpumask.h>
14 #include <linux/page-debug-flags.h>
15 #include <asm/page.h>
16 #include <asm/mmu.h>
17 
18 #ifndef AT_VECTOR_SIZE_ARCH
19 #define AT_VECTOR_SIZE_ARCH 0
20 #endif
21 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
22 
23 struct address_space;
24 
25 #define USE_SPLIT_PTLOCKS	(NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
26 
27 /*
28  * Each physical page in the system has a struct page associated with
29  * it to keep track of whatever it is we are using the page for at the
30  * moment. Note that we have no way to track which tasks are using
31  * a page, though if it is a pagecache page, rmap structures can tell us
32  * who is mapping it.
33  *
34  * The objects in struct page are organized in double word blocks in
35  * order to allows us to use atomic double word operations on portions
36  * of struct page. That is currently only used by slub but the arrangement
37  * allows the use of atomic double word operations on the flags/mapping
38  * and lru list pointers also.
39  */
40 struct page {
41 	/* First double word block */
42 	unsigned long flags;		/* Atomic flags, some possibly
43 					 * updated asynchronously */
44 	struct address_space *mapping;	/* If low bit clear, points to
45 					 * inode address_space, or NULL.
46 					 * If page mapped as anonymous
47 					 * memory, low bit is set, and
48 					 * it points to anon_vma object:
49 					 * see PAGE_MAPPING_ANON below.
50 					 */
51 	/* Second double word */
52 	struct {
53 		union {
54 			pgoff_t index;		/* Our offset within mapping. */
55 			void *freelist;		/* slub first free object */
56 		};
57 
58 		union {
59 			/* Used for cmpxchg_double in slub */
60 			unsigned long counters;
61 
62 			struct {
63 
64 				union {
65 					atomic_t _mapcount;	/* Count of ptes mapped in mms,
66 							 * to show when page is mapped
67 							 * & limit reverse map searches.
68 							 */
69 
70 					struct {
71 						unsigned inuse:16;
72 						unsigned objects:15;
73 						unsigned frozen:1;
74 					};
75 				};
76 				atomic_t _count;		/* Usage count, see below. */
77 			};
78 		};
79 	};
80 
81 	/* Third double word block */
82 	union {
83 		struct list_head lru;	/* Pageout list, eg. active_list
84 					 * protected by zone->lru_lock !
85 					 */
86 		struct {		/* slub per cpu partial pages */
87 			struct page *next;	/* Next partial slab */
88 #ifdef CONFIG_64BIT
89 			int pages;	/* Nr of partial slabs left */
90 			int pobjects;	/* Approximate # of objects */
91 #else
92 			short int pages;
93 			short int pobjects;
94 #endif
95 		};
96 	};
97 
98 	/* Remainder is not double word aligned */
99 	union {
100 		unsigned long private;		/* Mapping-private opaque data:
101 					 	 * usually used for buffer_heads
102 						 * if PagePrivate set; used for
103 						 * swp_entry_t if PageSwapCache;
104 						 * indicates order in the buddy
105 						 * system if PG_buddy is set.
106 						 */
107 #if USE_SPLIT_PTLOCKS
108 		spinlock_t ptl;
109 #endif
110 		struct kmem_cache *slab;	/* SLUB: Pointer to slab */
111 		struct page *first_page;	/* Compound tail pages */
112 	};
113 
114 	/*
115 	 * On machines where all RAM is mapped into kernel address space,
116 	 * we can simply calculate the virtual address. On machines with
117 	 * highmem some memory is mapped into kernel virtual memory
118 	 * dynamically, so we need a place to store that address.
119 	 * Note that this field could be 16 bits on x86 ... ;)
120 	 *
121 	 * Architectures with slow multiplication can define
122 	 * WANT_PAGE_VIRTUAL in asm/page.h
123 	 */
124 #if defined(WANT_PAGE_VIRTUAL)
125 	void *virtual;			/* Kernel virtual address (NULL if
126 					   not kmapped, ie. highmem) */
127 #endif /* WANT_PAGE_VIRTUAL */
128 #ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS
129 	unsigned long debug_flags;	/* Use atomic bitops on this */
130 #endif
131 
132 #ifdef CONFIG_KMEMCHECK
133 	/*
134 	 * kmemcheck wants to track the status of each byte in a page; this
135 	 * is a pointer to such a status block. NULL if not tracked.
136 	 */
137 	void *shadow;
138 #endif
139 }
140 /*
141  * If another subsystem starts using the double word pairing for atomic
142  * operations on struct page then it must change the #if to ensure
143  * proper alignment of the page struct.
144  */
145 #if defined(CONFIG_SLUB) && defined(CONFIG_CMPXCHG_LOCAL)
146 	__attribute__((__aligned__(2*sizeof(unsigned long))))
147 #endif
148 ;
149 
150 struct page_frag {
151 	struct page *page;
152 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
153 	__u32 offset;
154 	__u32 size;
155 #else
156 	__u16 offset;
157 	__u16 size;
158 #endif
159 };
160 
161 typedef unsigned long __nocast vm_flags_t;
162 
163 /*
164  * A region containing a mapping of a non-memory backed file under NOMMU
165  * conditions.  These are held in a global tree and are pinned by the VMAs that
166  * map parts of them.
167  */
168 struct vm_region {
169 	struct rb_node	vm_rb;		/* link in global region tree */
170 	vm_flags_t	vm_flags;	/* VMA vm_flags */
171 	unsigned long	vm_start;	/* start address of region */
172 	unsigned long	vm_end;		/* region initialised to here */
173 	unsigned long	vm_top;		/* region allocated to here */
174 	unsigned long	vm_pgoff;	/* the offset in vm_file corresponding to vm_start */
175 	struct file	*vm_file;	/* the backing file or NULL */
176 
177 	int		vm_usage;	/* region usage count (access under nommu_region_sem) */
178 	bool		vm_icache_flushed : 1; /* true if the icache has been flushed for
179 						* this region */
180 };
181 
182 /*
183  * This struct defines a memory VMM memory area. There is one of these
184  * per VM-area/task.  A VM area is any part of the process virtual memory
185  * space that has a special rule for the page-fault handlers (ie a shared
186  * library, the executable area etc).
187  */
188 struct vm_area_struct {
189 	struct mm_struct * vm_mm;	/* The address space we belong to. */
190 	unsigned long vm_start;		/* Our start address within vm_mm. */
191 	unsigned long vm_end;		/* The first byte after our end address
192 					   within vm_mm. */
193 
194 	/* linked list of VM areas per task, sorted by address */
195 	struct vm_area_struct *vm_next, *vm_prev;
196 
197 	pgprot_t vm_page_prot;		/* Access permissions of this VMA. */
198 	unsigned long vm_flags;		/* Flags, see mm.h. */
199 
200 	struct rb_node vm_rb;
201 
202 	/*
203 	 * For areas with an address space and backing store,
204 	 * linkage into the address_space->i_mmap prio tree, or
205 	 * linkage to the list of like vmas hanging off its node, or
206 	 * linkage of vma in the address_space->i_mmap_nonlinear list.
207 	 */
208 	union {
209 		struct {
210 			struct list_head list;
211 			void *parent;	/* aligns with prio_tree_node parent */
212 			struct vm_area_struct *head;
213 		} vm_set;
214 
215 		struct raw_prio_tree_node prio_tree_node;
216 	} shared;
217 
218 	/*
219 	 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
220 	 * list, after a COW of one of the file pages.	A MAP_SHARED vma
221 	 * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
222 	 * or brk vma (with NULL file) can only be in an anon_vma list.
223 	 */
224 	struct list_head anon_vma_chain; /* Serialized by mmap_sem &
225 					  * page_table_lock */
226 	struct anon_vma *anon_vma;	/* Serialized by page_table_lock */
227 
228 	/* Function pointers to deal with this struct. */
229 	const struct vm_operations_struct *vm_ops;
230 
231 	/* Information about our backing store: */
232 	unsigned long vm_pgoff;		/* Offset (within vm_file) in PAGE_SIZE
233 					   units, *not* PAGE_CACHE_SIZE */
234 	struct file * vm_file;		/* File we map to (can be NULL). */
235 	void * vm_private_data;		/* was vm_pte (shared mem) */
236 
237 #ifndef CONFIG_MMU
238 	struct vm_region *vm_region;	/* NOMMU mapping region */
239 #endif
240 #ifdef CONFIG_NUMA
241 	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
242 #endif
243 };
244 
245 struct core_thread {
246 	struct task_struct *task;
247 	struct core_thread *next;
248 };
249 
250 struct core_state {
251 	atomic_t nr_threads;
252 	struct core_thread dumper;
253 	struct completion startup;
254 };
255 
256 enum {
257 	MM_FILEPAGES,
258 	MM_ANONPAGES,
259 	MM_SWAPENTS,
260 	NR_MM_COUNTERS
261 };
262 
263 #if USE_SPLIT_PTLOCKS && defined(CONFIG_MMU)
264 #define SPLIT_RSS_COUNTING
265 /* per-thread cached information, */
266 struct task_rss_stat {
267 	int events;	/* for synchronization threshold */
268 	int count[NR_MM_COUNTERS];
269 };
270 #endif /* USE_SPLIT_PTLOCKS */
271 
272 struct mm_rss_stat {
273 	atomic_long_t count[NR_MM_COUNTERS];
274 };
275 
276 struct mm_struct {
277 	struct vm_area_struct * mmap;		/* list of VMAs */
278 	struct rb_root mm_rb;
279 	struct vm_area_struct * mmap_cache;	/* last find_vma result */
280 #ifdef CONFIG_MMU
281 	unsigned long (*get_unmapped_area) (struct file *filp,
282 				unsigned long addr, unsigned long len,
283 				unsigned long pgoff, unsigned long flags);
284 	void (*unmap_area) (struct mm_struct *mm, unsigned long addr);
285 #endif
286 	unsigned long mmap_base;		/* base of mmap area */
287 	unsigned long task_size;		/* size of task vm space */
288 	unsigned long cached_hole_size; 	/* if non-zero, the largest hole below free_area_cache */
289 	unsigned long free_area_cache;		/* first hole of size cached_hole_size or larger */
290 	pgd_t * pgd;
291 	atomic_t mm_users;			/* How many users with user space? */
292 	atomic_t mm_count;			/* How many references to "struct mm_struct" (users count as 1) */
293 	int map_count;				/* number of VMAs */
294 
295 	spinlock_t page_table_lock;		/* Protects page tables and some counters */
296 	struct rw_semaphore mmap_sem;
297 
298 	struct list_head mmlist;		/* List of maybe swapped mm's.	These are globally strung
299 						 * together off init_mm.mmlist, and are protected
300 						 * by mmlist_lock
301 						 */
302 
303 
304 	unsigned long hiwater_rss;	/* High-watermark of RSS usage */
305 	unsigned long hiwater_vm;	/* High-water virtual memory usage */
306 
307 	unsigned long total_vm, locked_vm, shared_vm, exec_vm;
308 	unsigned long stack_vm, reserved_vm, def_flags, nr_ptes;
309 	unsigned long start_code, end_code, start_data, end_data;
310 	unsigned long start_brk, brk, start_stack;
311 	unsigned long arg_start, arg_end, env_start, env_end;
312 
313 	unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
314 
315 	/*
316 	 * Special counters, in some configurations protected by the
317 	 * page_table_lock, in other configurations by being atomic.
318 	 */
319 	struct mm_rss_stat rss_stat;
320 
321 	struct linux_binfmt *binfmt;
322 
323 	cpumask_var_t cpu_vm_mask_var;
324 
325 	/* Architecture-specific MM context */
326 	mm_context_t context;
327 
328 	/* Swap token stuff */
329 	/*
330 	 * Last value of global fault stamp as seen by this process.
331 	 * In other words, this value gives an indication of how long
332 	 * it has been since this task got the token.
333 	 * Look at mm/thrash.c
334 	 */
335 	unsigned int faultstamp;
336 	unsigned int token_priority;
337 	unsigned int last_interval;
338 
339 	/* How many tasks sharing this mm are OOM_DISABLE */
340 	atomic_t oom_disable_count;
341 
342 	unsigned long flags; /* Must use atomic bitops to access the bits */
343 
344 	struct core_state *core_state; /* coredumping support */
345 #ifdef CONFIG_AIO
346 	spinlock_t		ioctx_lock;
347 	struct hlist_head	ioctx_list;
348 #endif
349 #ifdef CONFIG_MM_OWNER
350 	/*
351 	 * "owner" points to a task that is regarded as the canonical
352 	 * user/owner of this mm. All of the following must be true in
353 	 * order for it to be changed:
354 	 *
355 	 * current == mm->owner
356 	 * current->mm != mm
357 	 * new_owner->mm == mm
358 	 * new_owner->alloc_lock is held
359 	 */
360 	struct task_struct __rcu *owner;
361 #endif
362 
363 	/* store ref to file /proc/<pid>/exe symlink points to */
364 	struct file *exe_file;
365 	unsigned long num_exe_file_vmas;
366 #ifdef CONFIG_MMU_NOTIFIER
367 	struct mmu_notifier_mm *mmu_notifier_mm;
368 #endif
369 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
370 	pgtable_t pmd_huge_pte; /* protected by page_table_lock */
371 #endif
372 #ifdef CONFIG_CPUMASK_OFFSTACK
373 	struct cpumask cpumask_allocation;
374 #endif
375 };
376 
377 static inline void mm_init_cpumask(struct mm_struct *mm)
378 {
379 #ifdef CONFIG_CPUMASK_OFFSTACK
380 	mm->cpu_vm_mask_var = &mm->cpumask_allocation;
381 #endif
382 }
383 
384 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
385 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
386 {
387 	return mm->cpu_vm_mask_var;
388 }
389 
390 #endif /* _LINUX_MM_TYPES_H */
391