xref: /linux-6.15/include/linux/vmalloc.h (revision f7ee1f13)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_VMALLOC_H
3 #define _LINUX_VMALLOC_H
4 
5 #include <linux/spinlock.h>
6 #include <linux/init.h>
7 #include <linux/list.h>
8 #include <linux/llist.h>
9 #include <asm/page.h>		/* pgprot_t */
10 #include <linux/rbtree.h>
11 #include <linux/overflow.h>
12 
13 #include <asm/vmalloc.h>
14 
15 struct vm_area_struct;		/* vma defining user mapping in mm_types.h */
16 struct notifier_block;		/* in notifier.h */
17 
18 /* bits in flags of vmalloc's vm_struct below */
19 #define VM_IOREMAP		0x00000001	/* ioremap() and friends */
20 #define VM_ALLOC		0x00000002	/* vmalloc() */
21 #define VM_MAP			0x00000004	/* vmap()ed pages */
22 #define VM_USERMAP		0x00000008	/* suitable for remap_vmalloc_range */
23 #define VM_DMA_COHERENT		0x00000010	/* dma_alloc_coherent */
24 #define VM_UNINITIALIZED	0x00000020	/* vm_struct is not fully initialized */
25 #define VM_NO_GUARD		0x00000040      /* don't add guard page */
26 #define VM_KASAN		0x00000080      /* has allocated kasan shadow memory */
27 #define VM_FLUSH_RESET_PERMS	0x00000100	/* reset direct map and flush TLB on unmap, can't be freed in atomic context */
28 #define VM_MAP_PUT_PAGES	0x00000200	/* put pages and free array in vfree */
29 #define VM_NO_HUGE_VMAP		0x00000400	/* force PAGE_SIZE pte mapping */
30 
31 /*
32  * VM_KASAN is used slighly differently depending on CONFIG_KASAN_VMALLOC.
33  *
34  * If IS_ENABLED(CONFIG_KASAN_VMALLOC), VM_KASAN is set on a vm_struct after
35  * shadow memory has been mapped. It's used to handle allocation errors so that
36  * we don't try to poison shadow on free if it was never allocated.
37  *
38  * Otherwise, VM_KASAN is set for kasan_module_alloc() allocations and used to
39  * determine which allocations need the module shadow freed.
40  */
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 vmap_area_root)
77 	 *    2) or "busy" tree (root is free_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 };
84 
85 /* archs that select HAVE_ARCH_HUGE_VMAP should override one or more of these */
86 #ifndef arch_vmap_p4d_supported
87 static inline bool arch_vmap_p4d_supported(pgprot_t prot)
88 {
89 	return false;
90 }
91 #endif
92 
93 #ifndef arch_vmap_pud_supported
94 static inline bool arch_vmap_pud_supported(pgprot_t prot)
95 {
96 	return false;
97 }
98 #endif
99 
100 #ifndef arch_vmap_pmd_supported
101 static inline bool arch_vmap_pmd_supported(pgprot_t prot)
102 {
103 	return false;
104 }
105 #endif
106 
107 #ifndef arch_vmap_pte_range_map_size
108 static inline unsigned long arch_vmap_pte_range_map_size(unsigned long addr, unsigned long end,
109 							 u64 pfn, unsigned int max_page_shift)
110 {
111 	return PAGE_SIZE;
112 }
113 #endif
114 
115 /*
116  *	Highlevel APIs for driver use
117  */
118 extern void vm_unmap_ram(const void *mem, unsigned int count);
119 extern void *vm_map_ram(struct page **pages, unsigned int count, int node);
120 extern void vm_unmap_aliases(void);
121 
122 #ifdef CONFIG_MMU
123 extern void __init vmalloc_init(void);
124 extern unsigned long vmalloc_nr_pages(void);
125 #else
126 static inline void vmalloc_init(void)
127 {
128 }
129 static inline unsigned long vmalloc_nr_pages(void) { return 0; }
130 #endif
131 
132 extern void *vmalloc(unsigned long size);
133 extern void *vzalloc(unsigned long size);
134 extern void *vmalloc_user(unsigned long size);
135 extern void *vmalloc_node(unsigned long size, int node);
136 extern void *vzalloc_node(unsigned long size, int node);
137 extern void *vmalloc_32(unsigned long size);
138 extern void *vmalloc_32_user(unsigned long size);
139 extern void *__vmalloc(unsigned long size, gfp_t gfp_mask);
140 extern void *__vmalloc_node_range(unsigned long size, unsigned long align,
141 			unsigned long start, unsigned long end, gfp_t gfp_mask,
142 			pgprot_t prot, unsigned long vm_flags, int node,
143 			const void *caller);
144 void *__vmalloc_node(unsigned long size, unsigned long align, gfp_t gfp_mask,
145 		int node, const void *caller);
146 void *vmalloc_no_huge(unsigned long size);
147 
148 extern void vfree(const void *addr);
149 extern void vfree_atomic(const void *addr);
150 
151 extern void *vmap(struct page **pages, unsigned int count,
152 			unsigned long flags, pgprot_t prot);
153 void *vmap_pfn(unsigned long *pfns, unsigned int count, pgprot_t prot);
154 extern void vunmap(const void *addr);
155 
156 extern int remap_vmalloc_range_partial(struct vm_area_struct *vma,
157 				       unsigned long uaddr, void *kaddr,
158 				       unsigned long pgoff, unsigned long size);
159 
160 extern int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
161 							unsigned long pgoff);
162 
163 /*
164  * Architectures can set this mask to a combination of PGTBL_P?D_MODIFIED values
165  * and let generic vmalloc and ioremap code know when arch_sync_kernel_mappings()
166  * needs to be called.
167  */
168 #ifndef ARCH_PAGE_TABLE_SYNC_MASK
169 #define ARCH_PAGE_TABLE_SYNC_MASK 0
170 #endif
171 
172 /*
173  * There is no default implementation for arch_sync_kernel_mappings(). It is
174  * relied upon the compiler to optimize calls out if ARCH_PAGE_TABLE_SYNC_MASK
175  * is 0.
176  */
177 void arch_sync_kernel_mappings(unsigned long start, unsigned long end);
178 
179 /*
180  *	Lowlevel-APIs (not for driver use!)
181  */
182 
183 static inline size_t get_vm_area_size(const struct vm_struct *area)
184 {
185 	if (!(area->flags & VM_NO_GUARD))
186 		/* return actual size without guard page */
187 		return area->size - PAGE_SIZE;
188 	else
189 		return area->size;
190 
191 }
192 
193 extern struct vm_struct *get_vm_area(unsigned long size, unsigned long flags);
194 extern struct vm_struct *get_vm_area_caller(unsigned long size,
195 					unsigned long flags, const void *caller);
196 extern struct vm_struct *__get_vm_area_caller(unsigned long size,
197 					unsigned long flags,
198 					unsigned long start, unsigned long end,
199 					const void *caller);
200 void free_vm_area(struct vm_struct *area);
201 extern struct vm_struct *remove_vm_area(const void *addr);
202 extern struct vm_struct *find_vm_area(const void *addr);
203 
204 static inline bool is_vm_area_hugepages(const void *addr)
205 {
206 	/*
207 	 * This may not 100% tell if the area is mapped with > PAGE_SIZE
208 	 * page table entries, if for some reason the architecture indicates
209 	 * larger sizes are available but decides not to use them, nothing
210 	 * prevents that. This only indicates the size of the physical page
211 	 * allocated in the vmalloc layer.
212 	 */
213 #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC
214 	return find_vm_area(addr)->page_order > 0;
215 #else
216 	return false;
217 #endif
218 }
219 
220 #ifdef CONFIG_MMU
221 int vmap_range(unsigned long addr, unsigned long end,
222 			phys_addr_t phys_addr, pgprot_t prot,
223 			unsigned int max_page_shift);
224 void vunmap_range(unsigned long addr, unsigned long end);
225 static inline void set_vm_flush_reset_perms(void *addr)
226 {
227 	struct vm_struct *vm = find_vm_area(addr);
228 
229 	if (vm)
230 		vm->flags |= VM_FLUSH_RESET_PERMS;
231 }
232 
233 #else
234 static inline void set_vm_flush_reset_perms(void *addr)
235 {
236 }
237 #endif
238 
239 /* for /proc/kcore */
240 extern long vread(char *buf, char *addr, unsigned long count);
241 
242 /*
243  *	Internals.  Dont't use..
244  */
245 extern struct list_head vmap_area_list;
246 extern __init void vm_area_add_early(struct vm_struct *vm);
247 extern __init void vm_area_register_early(struct vm_struct *vm, size_t align);
248 
249 #ifdef CONFIG_SMP
250 # ifdef CONFIG_MMU
251 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
252 				     const size_t *sizes, int nr_vms,
253 				     size_t align);
254 
255 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms);
256 # else
257 static inline struct vm_struct **
258 pcpu_get_vm_areas(const unsigned long *offsets,
259 		const size_t *sizes, int nr_vms,
260 		size_t align)
261 {
262 	return NULL;
263 }
264 
265 static inline void
266 pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
267 {
268 }
269 # endif
270 #endif
271 
272 #ifdef CONFIG_MMU
273 #define VMALLOC_TOTAL (VMALLOC_END - VMALLOC_START)
274 #else
275 #define VMALLOC_TOTAL 0UL
276 #endif
277 
278 int register_vmap_purge_notifier(struct notifier_block *nb);
279 int unregister_vmap_purge_notifier(struct notifier_block *nb);
280 
281 #if defined(CONFIG_MMU) && defined(CONFIG_PRINTK)
282 bool vmalloc_dump_obj(void *object);
283 #else
284 static inline bool vmalloc_dump_obj(void *object) { return false; }
285 #endif
286 
287 #endif /* _LINUX_VMALLOC_H */
288