xref: /linux-6.15/include/linux/vmalloc.h (revision beeb9220)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_VMALLOC_H
3 #define _LINUX_VMALLOC_H
4 
5 #include <linux/alloc_tag.h>
6 #include <linux/sched.h>
7 #include <linux/spinlock.h>
8 #include <linux/init.h>
9 #include <linux/list.h>
10 #include <linux/llist.h>
11 #include <asm/page.h>		/* pgprot_t */
12 #include <linux/rbtree.h>
13 #include <linux/overflow.h>
14 
15 #include <asm/vmalloc.h>
16 
17 struct vm_area_struct;		/* vma defining user mapping in mm_types.h */
18 struct notifier_block;		/* in notifier.h */
19 struct iov_iter;		/* in uio.h */
20 
21 /* bits in flags of vmalloc's vm_struct below */
22 #define VM_IOREMAP		0x00000001	/* ioremap() and friends */
23 #define VM_ALLOC		0x00000002	/* vmalloc() */
24 #define VM_MAP			0x00000004	/* vmap()ed pages */
25 #define VM_USERMAP		0x00000008	/* suitable for remap_vmalloc_range */
26 #define VM_DMA_COHERENT		0x00000010	/* dma_alloc_coherent */
27 #define VM_UNINITIALIZED	0x00000020	/* vm_struct is not fully initialized */
28 #define VM_NO_GUARD		0x00000040      /* ***DANGEROUS*** don't add guard page */
29 #define VM_KASAN		0x00000080      /* has allocated kasan shadow memory */
30 #define VM_FLUSH_RESET_PERMS	0x00000100	/* reset direct map and flush TLB on unmap, can't be freed in atomic context */
31 #define VM_MAP_PUT_PAGES	0x00000200	/* put pages and free array in vfree */
32 #define VM_ALLOW_HUGE_VMAP	0x00000400      /* Allow for huge pages on archs with HAVE_ARCH_HUGE_VMALLOC */
33 
34 #if (defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)) && \
35 	!defined(CONFIG_KASAN_VMALLOC)
36 #define VM_DEFER_KMEMLEAK	0x00000800	/* defer kmemleak object creation */
37 #else
38 #define VM_DEFER_KMEMLEAK	0
39 #endif
40 #define VM_SPARSE		0x00001000	/* sparse vm_area. not all pages are present. */
41 
42 /* bits [20..32] reserved for arch specific ioremap internals */
43 
44 /*
45  * Maximum alignment for ioremap() regions.
46  * Can be overridden by arch-specific value.
47  */
48 #ifndef IOREMAP_MAX_ORDER
49 #define IOREMAP_MAX_ORDER	(7 + PAGE_SHIFT)	/* 128 pages */
50 #endif
51 
52 struct vm_struct {
53 	struct vm_struct	*next;
54 	void			*addr;
55 	unsigned long		size;
56 	unsigned long		flags;
57 	struct page		**pages;
58 #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC
59 	unsigned int		page_order;
60 #endif
61 	unsigned int		nr_pages;
62 	phys_addr_t		phys_addr;
63 	const void		*caller;
64 };
65 
66 struct vmap_area {
67 	unsigned long va_start;
68 	unsigned long va_end;
69 
70 	struct rb_node rb_node;         /* address sorted rbtree */
71 	struct list_head list;          /* address sorted list */
72 
73 	/*
74 	 * The following two variables can be packed, because
75 	 * a vmap_area object can be either:
76 	 *    1) in "free" tree (root is free_vmap_area_root)
77 	 *    2) or "busy" tree (root is vmap_area_root)
78 	 */
79 	union {
80 		unsigned long subtree_max_size; /* in "free" tree */
81 		struct vm_struct *vm;           /* in "busy" tree */
82 	};
83 	unsigned long flags; /* mark type of vm_map_ram area */
84 };
85 
86 /* archs that select HAVE_ARCH_HUGE_VMAP should override one or more of these */
87 #ifndef arch_vmap_p4d_supported
88 static inline bool arch_vmap_p4d_supported(pgprot_t prot)
89 {
90 	return false;
91 }
92 #endif
93 
94 #ifndef arch_vmap_pud_supported
95 static inline bool arch_vmap_pud_supported(pgprot_t prot)
96 {
97 	return false;
98 }
99 #endif
100 
101 #ifndef arch_vmap_pmd_supported
102 static inline bool arch_vmap_pmd_supported(pgprot_t prot)
103 {
104 	return false;
105 }
106 #endif
107 
108 #ifndef arch_vmap_pte_range_map_size
109 static inline unsigned long arch_vmap_pte_range_map_size(unsigned long addr, unsigned long end,
110 							 u64 pfn, unsigned int max_page_shift)
111 {
112 	return PAGE_SIZE;
113 }
114 #endif
115 
116 #ifndef arch_vmap_pte_supported_shift
117 static inline int arch_vmap_pte_supported_shift(unsigned long size)
118 {
119 	return PAGE_SHIFT;
120 }
121 #endif
122 
123 #ifndef arch_vmap_pgprot_tagged
124 static inline pgprot_t arch_vmap_pgprot_tagged(pgprot_t prot)
125 {
126 	return prot;
127 }
128 #endif
129 
130 /*
131  *	Highlevel APIs for driver use
132  */
133 extern void vm_unmap_ram(const void *mem, unsigned int count);
134 extern void *vm_map_ram(struct page **pages, unsigned int count, int node);
135 extern void vm_unmap_aliases(void);
136 
137 extern void *vmalloc_noprof(unsigned long size) __alloc_size(1);
138 #define vmalloc(...)		alloc_hooks(vmalloc_noprof(__VA_ARGS__))
139 
140 extern void *vzalloc_noprof(unsigned long size) __alloc_size(1);
141 #define vzalloc(...)		alloc_hooks(vzalloc_noprof(__VA_ARGS__))
142 
143 extern void *vmalloc_user_noprof(unsigned long size) __alloc_size(1);
144 #define vmalloc_user(...)	alloc_hooks(vmalloc_user_noprof(__VA_ARGS__))
145 
146 extern void *vmalloc_node_noprof(unsigned long size, int node) __alloc_size(1);
147 #define vmalloc_node(...)	alloc_hooks(vmalloc_node_noprof(__VA_ARGS__))
148 
149 extern void *vzalloc_node_noprof(unsigned long size, int node) __alloc_size(1);
150 #define vzalloc_node(...)	alloc_hooks(vzalloc_node_noprof(__VA_ARGS__))
151 
152 extern void *vmalloc_32_noprof(unsigned long size) __alloc_size(1);
153 #define vmalloc_32(...)		alloc_hooks(vmalloc_32_noprof(__VA_ARGS__))
154 
155 extern void *vmalloc_32_user_noprof(unsigned long size) __alloc_size(1);
156 #define vmalloc_32_user(...)	alloc_hooks(vmalloc_32_user_noprof(__VA_ARGS__))
157 
158 extern void *__vmalloc_noprof(unsigned long size, gfp_t gfp_mask) __alloc_size(1);
159 #define __vmalloc(...)		alloc_hooks(__vmalloc_noprof(__VA_ARGS__))
160 
161 extern void *__vmalloc_node_range_noprof(unsigned long size, unsigned long align,
162 			unsigned long start, unsigned long end, gfp_t gfp_mask,
163 			pgprot_t prot, unsigned long vm_flags, int node,
164 			const void *caller) __alloc_size(1);
165 #define __vmalloc_node_range(...)	alloc_hooks(__vmalloc_node_range_noprof(__VA_ARGS__))
166 
167 void *__vmalloc_node_noprof(unsigned long size, unsigned long align, gfp_t gfp_mask,
168 		int node, const void *caller) __alloc_size(1);
169 #define __vmalloc_node(...)	alloc_hooks(__vmalloc_node_noprof(__VA_ARGS__))
170 
171 void *vmalloc_huge_noprof(unsigned long size, gfp_t gfp_mask) __alloc_size(1);
172 #define vmalloc_huge(...)	alloc_hooks(vmalloc_huge_noprof(__VA_ARGS__))
173 
174 extern void *__vmalloc_array_noprof(size_t n, size_t size, gfp_t flags) __alloc_size(1, 2);
175 #define __vmalloc_array(...)	alloc_hooks(__vmalloc_array_noprof(__VA_ARGS__))
176 
177 extern void *vmalloc_array_noprof(size_t n, size_t size) __alloc_size(1, 2);
178 #define vmalloc_array(...)	alloc_hooks(vmalloc_array_noprof(__VA_ARGS__))
179 
180 extern void *__vcalloc_noprof(size_t n, size_t size, gfp_t flags) __alloc_size(1, 2);
181 #define __vcalloc(...)		alloc_hooks(__vcalloc_noprof(__VA_ARGS__))
182 
183 extern void *vcalloc_noprof(size_t n, size_t size) __alloc_size(1, 2);
184 #define vcalloc(...)		alloc_hooks(vcalloc_noprof(__VA_ARGS__))
185 
186 void * __must_check vrealloc_noprof(const void *p, size_t size, gfp_t flags)
187 		__realloc_size(2);
188 #define vrealloc(...)		alloc_hooks(vrealloc_noprof(__VA_ARGS__))
189 
190 extern void vfree(const void *addr);
191 extern void vfree_atomic(const void *addr);
192 
193 extern void *vmap(struct page **pages, unsigned int count,
194 			unsigned long flags, pgprot_t prot);
195 void *vmap_pfn(unsigned long *pfns, unsigned int count, pgprot_t prot);
196 extern void vunmap(const void *addr);
197 
198 extern int remap_vmalloc_range_partial(struct vm_area_struct *vma,
199 				       unsigned long uaddr, void *kaddr,
200 				       unsigned long pgoff, unsigned long size);
201 
202 extern int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
203 							unsigned long pgoff);
204 
205 /*
206  * Architectures can set this mask to a combination of PGTBL_P?D_MODIFIED values
207  * and let generic vmalloc and ioremap code know when arch_sync_kernel_mappings()
208  * needs to be called.
209  */
210 #ifndef ARCH_PAGE_TABLE_SYNC_MASK
211 #define ARCH_PAGE_TABLE_SYNC_MASK 0
212 #endif
213 
214 /*
215  * There is no default implementation for arch_sync_kernel_mappings(). It is
216  * relied upon the compiler to optimize calls out if ARCH_PAGE_TABLE_SYNC_MASK
217  * is 0.
218  */
219 void arch_sync_kernel_mappings(unsigned long start, unsigned long end);
220 
221 /*
222  *	Lowlevel-APIs (not for driver use!)
223  */
224 
225 static inline size_t get_vm_area_size(const struct vm_struct *area)
226 {
227 	if (!(area->flags & VM_NO_GUARD))
228 		/* return actual size without guard page */
229 		return area->size - PAGE_SIZE;
230 	else
231 		return area->size;
232 
233 }
234 
235 extern struct vm_struct *get_vm_area(unsigned long size, unsigned long flags);
236 extern struct vm_struct *get_vm_area_caller(unsigned long size,
237 					unsigned long flags, const void *caller);
238 extern struct vm_struct *__get_vm_area_caller(unsigned long size,
239 					unsigned long flags,
240 					unsigned long start, unsigned long end,
241 					const void *caller);
242 void free_vm_area(struct vm_struct *area);
243 extern struct vm_struct *remove_vm_area(const void *addr);
244 extern struct vm_struct *find_vm_area(const void *addr);
245 struct vmap_area *find_vmap_area(unsigned long addr);
246 
247 static inline bool is_vm_area_hugepages(const void *addr)
248 {
249 	/*
250 	 * This may not 100% tell if the area is mapped with > PAGE_SIZE
251 	 * page table entries, if for some reason the architecture indicates
252 	 * larger sizes are available but decides not to use them, nothing
253 	 * prevents that. This only indicates the size of the physical page
254 	 * allocated in the vmalloc layer.
255 	 */
256 #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC
257 	return find_vm_area(addr)->page_order > 0;
258 #else
259 	return false;
260 #endif
261 }
262 
263 /* for /proc/kcore */
264 long vread_iter(struct iov_iter *iter, const char *addr, size_t count);
265 
266 /*
267  *	Internals.  Don't use..
268  */
269 __init void vm_area_add_early(struct vm_struct *vm);
270 __init void vm_area_register_early(struct vm_struct *vm, size_t align);
271 
272 int register_vmap_purge_notifier(struct notifier_block *nb);
273 int unregister_vmap_purge_notifier(struct notifier_block *nb);
274 
275 #ifdef CONFIG_MMU
276 #define VMALLOC_TOTAL (VMALLOC_END - VMALLOC_START)
277 
278 unsigned long vmalloc_nr_pages(void);
279 
280 int vm_area_map_pages(struct vm_struct *area, unsigned long start,
281 		      unsigned long end, struct page **pages);
282 void vm_area_unmap_pages(struct vm_struct *area, unsigned long start,
283 			 unsigned long end);
284 void vunmap_range(unsigned long addr, unsigned long end);
285 
286 static inline void set_vm_flush_reset_perms(void *addr)
287 {
288 	struct vm_struct *vm = find_vm_area(addr);
289 
290 	if (vm)
291 		vm->flags |= VM_FLUSH_RESET_PERMS;
292 }
293 #else  /* !CONFIG_MMU */
294 #define VMALLOC_TOTAL 0UL
295 
296 static inline unsigned long vmalloc_nr_pages(void) { return 0; }
297 static inline void set_vm_flush_reset_perms(void *addr) {}
298 #endif /* CONFIG_MMU */
299 
300 #if defined(CONFIG_MMU) && defined(CONFIG_SMP)
301 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
302 				     const size_t *sizes, int nr_vms,
303 				     size_t align);
304 
305 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms);
306 # else
307 static inline struct vm_struct **
308 pcpu_get_vm_areas(const unsigned long *offsets,
309 		const size_t *sizes, int nr_vms,
310 		size_t align)
311 {
312 	return NULL;
313 }
314 
315 static inline void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) {}
316 #endif
317 
318 #if defined(CONFIG_MMU) && defined(CONFIG_PRINTK)
319 bool vmalloc_dump_obj(void *object);
320 #else
321 static inline bool vmalloc_dump_obj(void *object) { return false; }
322 #endif
323 
324 #endif /* _LINUX_VMALLOC_H */
325