xref: /linux-6.15/include/linux/mm_types.h (revision e978aa7d)
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 	struct list_head lru;		/* Pageout list, eg. active_list
83 					 * protected by zone->lru_lock !
84 					 */
85 
86 	/* Remainder is not double word aligned */
87 	union {
88 		unsigned long private;		/* Mapping-private opaque data:
89 					 	 * usually used for buffer_heads
90 						 * if PagePrivate set; used for
91 						 * swp_entry_t if PageSwapCache;
92 						 * indicates order in the buddy
93 						 * system if PG_buddy is set.
94 						 */
95 #if USE_SPLIT_PTLOCKS
96 		spinlock_t ptl;
97 #endif
98 		struct kmem_cache *slab;	/* SLUB: Pointer to slab */
99 		struct page *first_page;	/* Compound tail pages */
100 	};
101 
102 	/*
103 	 * On machines where all RAM is mapped into kernel address space,
104 	 * we can simply calculate the virtual address. On machines with
105 	 * highmem some memory is mapped into kernel virtual memory
106 	 * dynamically, so we need a place to store that address.
107 	 * Note that this field could be 16 bits on x86 ... ;)
108 	 *
109 	 * Architectures with slow multiplication can define
110 	 * WANT_PAGE_VIRTUAL in asm/page.h
111 	 */
112 #if defined(WANT_PAGE_VIRTUAL)
113 	void *virtual;			/* Kernel virtual address (NULL if
114 					   not kmapped, ie. highmem) */
115 #endif /* WANT_PAGE_VIRTUAL */
116 #ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS
117 	unsigned long debug_flags;	/* Use atomic bitops on this */
118 #endif
119 
120 #ifdef CONFIG_KMEMCHECK
121 	/*
122 	 * kmemcheck wants to track the status of each byte in a page; this
123 	 * is a pointer to such a status block. NULL if not tracked.
124 	 */
125 	void *shadow;
126 #endif
127 }
128 /*
129  * If another subsystem starts using the double word pairing for atomic
130  * operations on struct page then it must change the #if to ensure
131  * proper alignment of the page struct.
132  */
133 #if defined(CONFIG_SLUB) && defined(CONFIG_CMPXCHG_LOCAL)
134 	__attribute__((__aligned__(2*sizeof(unsigned long))))
135 #endif
136 ;
137 
138 typedef unsigned long __nocast vm_flags_t;
139 
140 /*
141  * A region containing a mapping of a non-memory backed file under NOMMU
142  * conditions.  These are held in a global tree and are pinned by the VMAs that
143  * map parts of them.
144  */
145 struct vm_region {
146 	struct rb_node	vm_rb;		/* link in global region tree */
147 	vm_flags_t	vm_flags;	/* VMA vm_flags */
148 	unsigned long	vm_start;	/* start address of region */
149 	unsigned long	vm_end;		/* region initialised to here */
150 	unsigned long	vm_top;		/* region allocated to here */
151 	unsigned long	vm_pgoff;	/* the offset in vm_file corresponding to vm_start */
152 	struct file	*vm_file;	/* the backing file or NULL */
153 
154 	int		vm_usage;	/* region usage count (access under nommu_region_sem) */
155 	bool		vm_icache_flushed : 1; /* true if the icache has been flushed for
156 						* this region */
157 };
158 
159 /*
160  * This struct defines a memory VMM memory area. There is one of these
161  * per VM-area/task.  A VM area is any part of the process virtual memory
162  * space that has a special rule for the page-fault handlers (ie a shared
163  * library, the executable area etc).
164  */
165 struct vm_area_struct {
166 	struct mm_struct * vm_mm;	/* The address space we belong to. */
167 	unsigned long vm_start;		/* Our start address within vm_mm. */
168 	unsigned long vm_end;		/* The first byte after our end address
169 					   within vm_mm. */
170 
171 	/* linked list of VM areas per task, sorted by address */
172 	struct vm_area_struct *vm_next, *vm_prev;
173 
174 	pgprot_t vm_page_prot;		/* Access permissions of this VMA. */
175 	unsigned long vm_flags;		/* Flags, see mm.h. */
176 
177 	struct rb_node vm_rb;
178 
179 	/*
180 	 * For areas with an address space and backing store,
181 	 * linkage into the address_space->i_mmap prio tree, or
182 	 * linkage to the list of like vmas hanging off its node, or
183 	 * linkage of vma in the address_space->i_mmap_nonlinear list.
184 	 */
185 	union {
186 		struct {
187 			struct list_head list;
188 			void *parent;	/* aligns with prio_tree_node parent */
189 			struct vm_area_struct *head;
190 		} vm_set;
191 
192 		struct raw_prio_tree_node prio_tree_node;
193 	} shared;
194 
195 	/*
196 	 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
197 	 * list, after a COW of one of the file pages.	A MAP_SHARED vma
198 	 * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
199 	 * or brk vma (with NULL file) can only be in an anon_vma list.
200 	 */
201 	struct list_head anon_vma_chain; /* Serialized by mmap_sem &
202 					  * page_table_lock */
203 	struct anon_vma *anon_vma;	/* Serialized by page_table_lock */
204 
205 	/* Function pointers to deal with this struct. */
206 	const struct vm_operations_struct *vm_ops;
207 
208 	/* Information about our backing store: */
209 	unsigned long vm_pgoff;		/* Offset (within vm_file) in PAGE_SIZE
210 					   units, *not* PAGE_CACHE_SIZE */
211 	struct file * vm_file;		/* File we map to (can be NULL). */
212 	void * vm_private_data;		/* was vm_pte (shared mem) */
213 
214 #ifndef CONFIG_MMU
215 	struct vm_region *vm_region;	/* NOMMU mapping region */
216 #endif
217 #ifdef CONFIG_NUMA
218 	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
219 #endif
220 };
221 
222 struct core_thread {
223 	struct task_struct *task;
224 	struct core_thread *next;
225 };
226 
227 struct core_state {
228 	atomic_t nr_threads;
229 	struct core_thread dumper;
230 	struct completion startup;
231 };
232 
233 enum {
234 	MM_FILEPAGES,
235 	MM_ANONPAGES,
236 	MM_SWAPENTS,
237 	NR_MM_COUNTERS
238 };
239 
240 #if USE_SPLIT_PTLOCKS && defined(CONFIG_MMU)
241 #define SPLIT_RSS_COUNTING
242 /* per-thread cached information, */
243 struct task_rss_stat {
244 	int events;	/* for synchronization threshold */
245 	int count[NR_MM_COUNTERS];
246 };
247 #endif /* USE_SPLIT_PTLOCKS */
248 
249 struct mm_rss_stat {
250 	atomic_long_t count[NR_MM_COUNTERS];
251 };
252 
253 struct mm_struct {
254 	struct vm_area_struct * mmap;		/* list of VMAs */
255 	struct rb_root mm_rb;
256 	struct vm_area_struct * mmap_cache;	/* last find_vma result */
257 #ifdef CONFIG_MMU
258 	unsigned long (*get_unmapped_area) (struct file *filp,
259 				unsigned long addr, unsigned long len,
260 				unsigned long pgoff, unsigned long flags);
261 	void (*unmap_area) (struct mm_struct *mm, unsigned long addr);
262 #endif
263 	unsigned long mmap_base;		/* base of mmap area */
264 	unsigned long task_size;		/* size of task vm space */
265 	unsigned long cached_hole_size; 	/* if non-zero, the largest hole below free_area_cache */
266 	unsigned long free_area_cache;		/* first hole of size cached_hole_size or larger */
267 	pgd_t * pgd;
268 	atomic_t mm_users;			/* How many users with user space? */
269 	atomic_t mm_count;			/* How many references to "struct mm_struct" (users count as 1) */
270 	int map_count;				/* number of VMAs */
271 
272 	spinlock_t page_table_lock;		/* Protects page tables and some counters */
273 	struct rw_semaphore mmap_sem;
274 
275 	struct list_head mmlist;		/* List of maybe swapped mm's.	These are globally strung
276 						 * together off init_mm.mmlist, and are protected
277 						 * by mmlist_lock
278 						 */
279 
280 
281 	unsigned long hiwater_rss;	/* High-watermark of RSS usage */
282 	unsigned long hiwater_vm;	/* High-water virtual memory usage */
283 
284 	unsigned long total_vm, locked_vm, shared_vm, exec_vm;
285 	unsigned long stack_vm, reserved_vm, def_flags, nr_ptes;
286 	unsigned long start_code, end_code, start_data, end_data;
287 	unsigned long start_brk, brk, start_stack;
288 	unsigned long arg_start, arg_end, env_start, env_end;
289 
290 	unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
291 
292 	/*
293 	 * Special counters, in some configurations protected by the
294 	 * page_table_lock, in other configurations by being atomic.
295 	 */
296 	struct mm_rss_stat rss_stat;
297 
298 	struct linux_binfmt *binfmt;
299 
300 	cpumask_var_t cpu_vm_mask_var;
301 
302 	/* Architecture-specific MM context */
303 	mm_context_t context;
304 
305 	/* Swap token stuff */
306 	/*
307 	 * Last value of global fault stamp as seen by this process.
308 	 * In other words, this value gives an indication of how long
309 	 * it has been since this task got the token.
310 	 * Look at mm/thrash.c
311 	 */
312 	unsigned int faultstamp;
313 	unsigned int token_priority;
314 	unsigned int last_interval;
315 
316 	/* How many tasks sharing this mm are OOM_DISABLE */
317 	atomic_t oom_disable_count;
318 
319 	unsigned long flags; /* Must use atomic bitops to access the bits */
320 
321 	struct core_state *core_state; /* coredumping support */
322 #ifdef CONFIG_AIO
323 	spinlock_t		ioctx_lock;
324 	struct hlist_head	ioctx_list;
325 #endif
326 #ifdef CONFIG_MM_OWNER
327 	/*
328 	 * "owner" points to a task that is regarded as the canonical
329 	 * user/owner of this mm. All of the following must be true in
330 	 * order for it to be changed:
331 	 *
332 	 * current == mm->owner
333 	 * current->mm != mm
334 	 * new_owner->mm == mm
335 	 * new_owner->alloc_lock is held
336 	 */
337 	struct task_struct __rcu *owner;
338 #endif
339 
340 	/* store ref to file /proc/<pid>/exe symlink points to */
341 	struct file *exe_file;
342 	unsigned long num_exe_file_vmas;
343 #ifdef CONFIG_MMU_NOTIFIER
344 	struct mmu_notifier_mm *mmu_notifier_mm;
345 #endif
346 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
347 	pgtable_t pmd_huge_pte; /* protected by page_table_lock */
348 #endif
349 #ifdef CONFIG_CPUMASK_OFFSTACK
350 	struct cpumask cpumask_allocation;
351 #endif
352 };
353 
354 static inline void mm_init_cpumask(struct mm_struct *mm)
355 {
356 #ifdef CONFIG_CPUMASK_OFFSTACK
357 	mm->cpu_vm_mask_var = &mm->cpumask_allocation;
358 #endif
359 }
360 
361 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
362 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
363 {
364 	return mm->cpu_vm_mask_var;
365 }
366 
367 #endif /* _LINUX_MM_TYPES_H */
368