xref: /linux-6.15/include/linux/mm_types.h (revision 6fa79bca)
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 <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/slob first free object */
56 			bool pfmemalloc;	/* If set by the page allocator,
57 						 * ALLOC_NO_WATERMARKS was set
58 						 * and the low watermark was not
59 						 * met implying that the system
60 						 * is under some pressure. The
61 						 * caller should try ensure
62 						 * this page is only used to
63 						 * free other pages.
64 						 */
65 		};
66 
67 		union {
68 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
69 	defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
70 			/* Used for cmpxchg_double in slub */
71 			unsigned long counters;
72 #else
73 			/*
74 			 * Keep _count separate from slub cmpxchg_double data.
75 			 * As the rest of the double word is protected by
76 			 * slab_lock but _count is not.
77 			 */
78 			unsigned counters;
79 #endif
80 
81 			struct {
82 
83 				union {
84 					/*
85 					 * Count of ptes mapped in
86 					 * mms, to show when page is
87 					 * mapped & limit reverse map
88 					 * searches.
89 					 *
90 					 * Used also for tail pages
91 					 * refcounting instead of
92 					 * _count. Tail pages cannot
93 					 * be mapped and keeping the
94 					 * tail page _count zero at
95 					 * all times guarantees
96 					 * get_page_unless_zero() will
97 					 * never succeed on tail
98 					 * pages.
99 					 */
100 					atomic_t _mapcount;
101 
102 					struct { /* SLUB */
103 						unsigned inuse:16;
104 						unsigned objects:15;
105 						unsigned frozen:1;
106 					};
107 					int units;	/* SLOB */
108 				};
109 				atomic_t _count;		/* Usage count, see below. */
110 			};
111 		};
112 	};
113 
114 	/* Third double word block */
115 	union {
116 		struct list_head lru;	/* Pageout list, eg. active_list
117 					 * protected by zone->lru_lock !
118 					 */
119 		struct {		/* slub per cpu partial pages */
120 			struct page *next;	/* Next partial slab */
121 #ifdef CONFIG_64BIT
122 			int pages;	/* Nr of partial slabs left */
123 			int pobjects;	/* Approximate # of objects */
124 #else
125 			short int pages;
126 			short int pobjects;
127 #endif
128 		};
129 
130 		struct list_head list;	/* slobs list of pages */
131 		struct {		/* slab fields */
132 			struct kmem_cache *slab_cache;
133 			struct slab *slab_page;
134 		};
135 	};
136 
137 	/* Remainder is not double word aligned */
138 	union {
139 		unsigned long private;		/* Mapping-private opaque data:
140 					 	 * usually used for buffer_heads
141 						 * if PagePrivate set; used for
142 						 * swp_entry_t if PageSwapCache;
143 						 * indicates order in the buddy
144 						 * system if PG_buddy is set.
145 						 */
146 #if USE_SPLIT_PTLOCKS
147 		spinlock_t ptl;
148 #endif
149 		struct kmem_cache *slab;	/* SLUB: Pointer to slab */
150 		struct page *first_page;	/* Compound tail pages */
151 	};
152 
153 	/*
154 	 * On machines where all RAM is mapped into kernel address space,
155 	 * we can simply calculate the virtual address. On machines with
156 	 * highmem some memory is mapped into kernel virtual memory
157 	 * dynamically, so we need a place to store that address.
158 	 * Note that this field could be 16 bits on x86 ... ;)
159 	 *
160 	 * Architectures with slow multiplication can define
161 	 * WANT_PAGE_VIRTUAL in asm/page.h
162 	 */
163 #if defined(WANT_PAGE_VIRTUAL)
164 	void *virtual;			/* Kernel virtual address (NULL if
165 					   not kmapped, ie. highmem) */
166 #endif /* WANT_PAGE_VIRTUAL */
167 #ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS
168 	unsigned long debug_flags;	/* Use atomic bitops on this */
169 #endif
170 
171 #ifdef CONFIG_KMEMCHECK
172 	/*
173 	 * kmemcheck wants to track the status of each byte in a page; this
174 	 * is a pointer to such a status block. NULL if not tracked.
175 	 */
176 	void *shadow;
177 #endif
178 }
179 /*
180  * The struct page can be forced to be double word aligned so that atomic ops
181  * on double words work. The SLUB allocator can make use of such a feature.
182  */
183 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
184 	__aligned(2 * sizeof(unsigned long))
185 #endif
186 ;
187 
188 struct page_frag {
189 	struct page *page;
190 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
191 	__u32 offset;
192 	__u32 size;
193 #else
194 	__u16 offset;
195 	__u16 size;
196 #endif
197 };
198 
199 typedef unsigned long __nocast vm_flags_t;
200 
201 /*
202  * A region containing a mapping of a non-memory backed file under NOMMU
203  * conditions.  These are held in a global tree and are pinned by the VMAs that
204  * map parts of them.
205  */
206 struct vm_region {
207 	struct rb_node	vm_rb;		/* link in global region tree */
208 	vm_flags_t	vm_flags;	/* VMA vm_flags */
209 	unsigned long	vm_start;	/* start address of region */
210 	unsigned long	vm_end;		/* region initialised to here */
211 	unsigned long	vm_top;		/* region allocated to here */
212 	unsigned long	vm_pgoff;	/* the offset in vm_file corresponding to vm_start */
213 	struct file	*vm_file;	/* the backing file or NULL */
214 
215 	int		vm_usage;	/* region usage count (access under nommu_region_sem) */
216 	bool		vm_icache_flushed : 1; /* true if the icache has been flushed for
217 						* this region */
218 };
219 
220 /*
221  * This struct defines a memory VMM memory area. There is one of these
222  * per VM-area/task.  A VM area is any part of the process virtual memory
223  * space that has a special rule for the page-fault handlers (ie a shared
224  * library, the executable area etc).
225  */
226 struct vm_area_struct {
227 	struct mm_struct * vm_mm;	/* The address space we belong to. */
228 	unsigned long vm_start;		/* Our start address within vm_mm. */
229 	unsigned long vm_end;		/* The first byte after our end address
230 					   within vm_mm. */
231 
232 	/* linked list of VM areas per task, sorted by address */
233 	struct vm_area_struct *vm_next, *vm_prev;
234 
235 	pgprot_t vm_page_prot;		/* Access permissions of this VMA. */
236 	unsigned long vm_flags;		/* Flags, see mm.h. */
237 
238 	struct rb_node vm_rb;
239 
240 	/*
241 	 * For areas with an address space and backing store,
242 	 * linkage into the address_space->i_mmap interval tree, or
243 	 * linkage of vma in the address_space->i_mmap_nonlinear list.
244 	 */
245 	union {
246 		struct {
247 			struct rb_node rb;
248 			unsigned long rb_subtree_last;
249 		} linear;
250 		struct list_head nonlinear;
251 	} shared;
252 
253 	/*
254 	 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
255 	 * list, after a COW of one of the file pages.	A MAP_SHARED vma
256 	 * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
257 	 * or brk vma (with NULL file) can only be in an anon_vma list.
258 	 */
259 	struct list_head anon_vma_chain; /* Serialized by mmap_sem &
260 					  * page_table_lock */
261 	struct anon_vma *anon_vma;	/* Serialized by page_table_lock */
262 
263 	/* Function pointers to deal with this struct. */
264 	const struct vm_operations_struct *vm_ops;
265 
266 	/* Information about our backing store: */
267 	unsigned long vm_pgoff;		/* Offset (within vm_file) in PAGE_SIZE
268 					   units, *not* PAGE_CACHE_SIZE */
269 	struct file * vm_file;		/* File we map to (can be NULL). */
270 	void * vm_private_data;		/* was vm_pte (shared mem) */
271 
272 #ifndef CONFIG_MMU
273 	struct vm_region *vm_region;	/* NOMMU mapping region */
274 #endif
275 #ifdef CONFIG_NUMA
276 	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
277 #endif
278 };
279 
280 struct core_thread {
281 	struct task_struct *task;
282 	struct core_thread *next;
283 };
284 
285 struct core_state {
286 	atomic_t nr_threads;
287 	struct core_thread dumper;
288 	struct completion startup;
289 };
290 
291 enum {
292 	MM_FILEPAGES,
293 	MM_ANONPAGES,
294 	MM_SWAPENTS,
295 	NR_MM_COUNTERS
296 };
297 
298 #if USE_SPLIT_PTLOCKS && defined(CONFIG_MMU)
299 #define SPLIT_RSS_COUNTING
300 /* per-thread cached information, */
301 struct task_rss_stat {
302 	int events;	/* for synchronization threshold */
303 	int count[NR_MM_COUNTERS];
304 };
305 #endif /* USE_SPLIT_PTLOCKS */
306 
307 struct mm_rss_stat {
308 	atomic_long_t count[NR_MM_COUNTERS];
309 };
310 
311 struct mm_struct {
312 	struct vm_area_struct * mmap;		/* list of VMAs */
313 	struct rb_root mm_rb;
314 	struct vm_area_struct * mmap_cache;	/* last find_vma result */
315 #ifdef CONFIG_MMU
316 	unsigned long (*get_unmapped_area) (struct file *filp,
317 				unsigned long addr, unsigned long len,
318 				unsigned long pgoff, unsigned long flags);
319 	void (*unmap_area) (struct mm_struct *mm, unsigned long addr);
320 #endif
321 	unsigned long mmap_base;		/* base of mmap area */
322 	unsigned long task_size;		/* size of task vm space */
323 	unsigned long cached_hole_size; 	/* if non-zero, the largest hole below free_area_cache */
324 	unsigned long free_area_cache;		/* first hole of size cached_hole_size or larger */
325 	pgd_t * pgd;
326 	atomic_t mm_users;			/* How many users with user space? */
327 	atomic_t mm_count;			/* How many references to "struct mm_struct" (users count as 1) */
328 	int map_count;				/* number of VMAs */
329 
330 	spinlock_t page_table_lock;		/* Protects page tables and some counters */
331 	struct rw_semaphore mmap_sem;
332 
333 	struct list_head mmlist;		/* List of maybe swapped mm's.	These are globally strung
334 						 * together off init_mm.mmlist, and are protected
335 						 * by mmlist_lock
336 						 */
337 
338 
339 	unsigned long hiwater_rss;	/* High-watermark of RSS usage */
340 	unsigned long hiwater_vm;	/* High-water virtual memory usage */
341 
342 	unsigned long total_vm;		/* Total pages mapped */
343 	unsigned long locked_vm;	/* Pages that have PG_mlocked set */
344 	unsigned long pinned_vm;	/* Refcount permanently increased */
345 	unsigned long shared_vm;	/* Shared pages (files) */
346 	unsigned long exec_vm;		/* VM_EXEC & ~VM_WRITE */
347 	unsigned long stack_vm;		/* VM_GROWSUP/DOWN */
348 	unsigned long def_flags;
349 	unsigned long nr_ptes;		/* Page table pages */
350 	unsigned long start_code, end_code, start_data, end_data;
351 	unsigned long start_brk, brk, start_stack;
352 	unsigned long arg_start, arg_end, env_start, env_end;
353 
354 	unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
355 
356 	/*
357 	 * Special counters, in some configurations protected by the
358 	 * page_table_lock, in other configurations by being atomic.
359 	 */
360 	struct mm_rss_stat rss_stat;
361 
362 	struct linux_binfmt *binfmt;
363 
364 	cpumask_var_t cpu_vm_mask_var;
365 
366 	/* Architecture-specific MM context */
367 	mm_context_t context;
368 
369 	unsigned long flags; /* Must use atomic bitops to access the bits */
370 
371 	struct core_state *core_state; /* coredumping support */
372 #ifdef CONFIG_AIO
373 	spinlock_t		ioctx_lock;
374 	struct hlist_head	ioctx_list;
375 #endif
376 #ifdef CONFIG_MM_OWNER
377 	/*
378 	 * "owner" points to a task that is regarded as the canonical
379 	 * user/owner of this mm. All of the following must be true in
380 	 * order for it to be changed:
381 	 *
382 	 * current == mm->owner
383 	 * current->mm != mm
384 	 * new_owner->mm == mm
385 	 * new_owner->alloc_lock is held
386 	 */
387 	struct task_struct __rcu *owner;
388 #endif
389 
390 	/* store ref to file /proc/<pid>/exe symlink points to */
391 	struct file *exe_file;
392 #ifdef CONFIG_MMU_NOTIFIER
393 	struct mmu_notifier_mm *mmu_notifier_mm;
394 #endif
395 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
396 	pgtable_t pmd_huge_pte; /* protected by page_table_lock */
397 #endif
398 #ifdef CONFIG_CPUMASK_OFFSTACK
399 	struct cpumask cpumask_allocation;
400 #endif
401 	struct uprobes_state uprobes_state;
402 };
403 
404 static inline void mm_init_cpumask(struct mm_struct *mm)
405 {
406 #ifdef CONFIG_CPUMASK_OFFSTACK
407 	mm->cpu_vm_mask_var = &mm->cpumask_allocation;
408 #endif
409 }
410 
411 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
412 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
413 {
414 	return mm->cpu_vm_mask_var;
415 }
416 
417 #endif /* _LINUX_MM_TYPES_H */
418