1 #ifndef __ASM_SH_UNALIGNED_SH4A_H
2 #define __ASM_SH_UNALIGNED_SH4A_H
3 
4 /*
5  * SH-4A has support for unaligned 32-bit loads, and 32-bit loads only.
6  * Support for 64-bit accesses are done through shifting and masking
7  * relative to the endianness. Unaligned stores are not supported by the
8  * instruction encoding, so these continue to use the packed
9  * struct.
10  *
11  * The same note as with the movli.l/movco.l pair applies here, as long
12  * as the load is gauranteed to be inlined, nothing else will hook in to
13  * r0 and we get the return value for free.
14  *
15  * NOTE: Due to the fact we require r0 encoding, care should be taken to
16  * avoid mixing these heavily with other r0 consumers, such as the atomic
17  * ops. Failure to adhere to this can result in the compiler running out
18  * of spill registers and blowing up when building at low optimization
19  * levels. See http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34777.
20  */
21 #include <linux/types.h>
22 #include <asm/byteorder.h>
23 
__get_unaligned_cpu32(const u8 * p)24 static __always_inline u32 __get_unaligned_cpu32(const u8 *p)
25 {
26 	unsigned long unaligned;
27 
28 	__asm__ __volatile__ (
29 		"movua.l	@%1, %0\n\t"
30 		 : "=z" (unaligned)
31 		 : "r" (p)
32 	);
33 
34 	return unaligned;
35 }
36 
37 struct __una_u16 { u16 x __attribute__((packed)); };
38 struct __una_u32 { u32 x __attribute__((packed)); };
39 struct __una_u64 { u64 x __attribute__((packed)); };
40 
__get_unaligned_cpu16(const u8 * p)41 static inline u16 __get_unaligned_cpu16(const u8 *p)
42 {
43 #ifdef __LITTLE_ENDIAN
44 	return p[0] | p[1] << 8;
45 #else
46 	return p[0] << 8 | p[1];
47 #endif
48 }
49 
50 /*
51  * Even though movua.l supports auto-increment on the read side, it can
52  * only store to r0 due to instruction encoding constraints, so just let
53  * the compiler sort it out on its own.
54  */
__get_unaligned_cpu64(const u8 * p)55 static inline u64 __get_unaligned_cpu64(const u8 *p)
56 {
57 #ifdef __LITTLE_ENDIAN
58 	return (u64)__get_unaligned_cpu32(p + 4) << 32 |
59 		    __get_unaligned_cpu32(p);
60 #else
61 	return (u64)__get_unaligned_cpu32(p) << 32 |
62 		    __get_unaligned_cpu32(p + 4);
63 #endif
64 }
65 
get_unaligned_le16(const void * p)66 static inline u16 get_unaligned_le16(const void *p)
67 {
68 	return le16_to_cpu(__get_unaligned_cpu16(p));
69 }
70 
get_unaligned_le32(const void * p)71 static inline u32 get_unaligned_le32(const void *p)
72 {
73 	return le32_to_cpu(__get_unaligned_cpu32(p));
74 }
75 
get_unaligned_le64(const void * p)76 static inline u64 get_unaligned_le64(const void *p)
77 {
78 	return le64_to_cpu(__get_unaligned_cpu64(p));
79 }
80 
get_unaligned_be16(const void * p)81 static inline u16 get_unaligned_be16(const void *p)
82 {
83 	return be16_to_cpu(__get_unaligned_cpu16(p));
84 }
85 
get_unaligned_be32(const void * p)86 static inline u32 get_unaligned_be32(const void *p)
87 {
88 	return be32_to_cpu(__get_unaligned_cpu32(p));
89 }
90 
get_unaligned_be64(const void * p)91 static inline u64 get_unaligned_be64(const void *p)
92 {
93 	return be64_to_cpu(__get_unaligned_cpu64(p));
94 }
95 
__put_le16_noalign(u8 * p,u16 val)96 static inline void __put_le16_noalign(u8 *p, u16 val)
97 {
98 	*p++ = val;
99 	*p++ = val >> 8;
100 }
101 
__put_le32_noalign(u8 * p,u32 val)102 static inline void __put_le32_noalign(u8 *p, u32 val)
103 {
104 	__put_le16_noalign(p, val);
105 	__put_le16_noalign(p + 2, val >> 16);
106 }
107 
__put_le64_noalign(u8 * p,u64 val)108 static inline void __put_le64_noalign(u8 *p, u64 val)
109 {
110 	__put_le32_noalign(p, val);
111 	__put_le32_noalign(p + 4, val >> 32);
112 }
113 
__put_be16_noalign(u8 * p,u16 val)114 static inline void __put_be16_noalign(u8 *p, u16 val)
115 {
116 	*p++ = val >> 8;
117 	*p++ = val;
118 }
119 
__put_be32_noalign(u8 * p,u32 val)120 static inline void __put_be32_noalign(u8 *p, u32 val)
121 {
122 	__put_be16_noalign(p, val >> 16);
123 	__put_be16_noalign(p + 2, val);
124 }
125 
__put_be64_noalign(u8 * p,u64 val)126 static inline void __put_be64_noalign(u8 *p, u64 val)
127 {
128 	__put_be32_noalign(p, val >> 32);
129 	__put_be32_noalign(p + 4, val);
130 }
131 
put_unaligned_le16(u16 val,void * p)132 static inline void put_unaligned_le16(u16 val, void *p)
133 {
134 #ifdef __LITTLE_ENDIAN
135 	((struct __una_u16 *)p)->x = val;
136 #else
137 	__put_le16_noalign(p, val);
138 #endif
139 }
140 
put_unaligned_le32(u32 val,void * p)141 static inline void put_unaligned_le32(u32 val, void *p)
142 {
143 #ifdef __LITTLE_ENDIAN
144 	((struct __una_u32 *)p)->x = val;
145 #else
146 	__put_le32_noalign(p, val);
147 #endif
148 }
149 
put_unaligned_le64(u64 val,void * p)150 static inline void put_unaligned_le64(u64 val, void *p)
151 {
152 #ifdef __LITTLE_ENDIAN
153 	((struct __una_u64 *)p)->x = val;
154 #else
155 	__put_le64_noalign(p, val);
156 #endif
157 }
158 
put_unaligned_be16(u16 val,void * p)159 static inline void put_unaligned_be16(u16 val, void *p)
160 {
161 #ifdef __BIG_ENDIAN
162 	((struct __una_u16 *)p)->x = val;
163 #else
164 	__put_be16_noalign(p, val);
165 #endif
166 }
167 
put_unaligned_be32(u32 val,void * p)168 static inline void put_unaligned_be32(u32 val, void *p)
169 {
170 #ifdef __BIG_ENDIAN
171 	((struct __una_u32 *)p)->x = val;
172 #else
173 	__put_be32_noalign(p, val);
174 #endif
175 }
176 
put_unaligned_be64(u64 val,void * p)177 static inline void put_unaligned_be64(u64 val, void *p)
178 {
179 #ifdef __BIG_ENDIAN
180 	((struct __una_u64 *)p)->x = val;
181 #else
182 	__put_be64_noalign(p, val);
183 #endif
184 }
185 
186 /*
187  * Cause a link-time error if we try an unaligned access other than
188  * 1,2,4 or 8 bytes long
189  */
190 extern void __bad_unaligned_access_size(void);
191 
192 #define __get_unaligned_le(ptr) ((__force typeof(*(ptr)))({			\
193 	__builtin_choose_expr(sizeof(*(ptr)) == 1, *(ptr),			\
194 	__builtin_choose_expr(sizeof(*(ptr)) == 2, get_unaligned_le16((ptr)),	\
195 	__builtin_choose_expr(sizeof(*(ptr)) == 4, get_unaligned_le32((ptr)),	\
196 	__builtin_choose_expr(sizeof(*(ptr)) == 8, get_unaligned_le64((ptr)),	\
197 	__bad_unaligned_access_size()))));					\
198 	}))
199 
200 #define __get_unaligned_be(ptr) ((__force typeof(*(ptr)))({			\
201 	__builtin_choose_expr(sizeof(*(ptr)) == 1, *(ptr),			\
202 	__builtin_choose_expr(sizeof(*(ptr)) == 2, get_unaligned_be16((ptr)),	\
203 	__builtin_choose_expr(sizeof(*(ptr)) == 4, get_unaligned_be32((ptr)),	\
204 	__builtin_choose_expr(sizeof(*(ptr)) == 8, get_unaligned_be64((ptr)),	\
205 	__bad_unaligned_access_size()))));					\
206 	}))
207 
208 #define __put_unaligned_le(val, ptr) ({					\
209 	void *__gu_p = (ptr);						\
210 	switch (sizeof(*(ptr))) {					\
211 	case 1:								\
212 		*(u8 *)__gu_p = (__force u8)(val);			\
213 		break;							\
214 	case 2:								\
215 		put_unaligned_le16((__force u16)(val), __gu_p);		\
216 		break;							\
217 	case 4:								\
218 		put_unaligned_le32((__force u32)(val), __gu_p);		\
219 		break;							\
220 	case 8:								\
221 		put_unaligned_le64((__force u64)(val), __gu_p);		\
222 		break;							\
223 	default:							\
224 		__bad_unaligned_access_size();				\
225 		break;							\
226 	}								\
227 	(void)0; })
228 
229 #define __put_unaligned_be(val, ptr) ({					\
230 	void *__gu_p = (ptr);						\
231 	switch (sizeof(*(ptr))) {					\
232 	case 1:								\
233 		*(u8 *)__gu_p = (__force u8)(val);			\
234 		break;							\
235 	case 2:								\
236 		put_unaligned_be16((__force u16)(val), __gu_p);		\
237 		break;							\
238 	case 4:								\
239 		put_unaligned_be32((__force u32)(val), __gu_p);		\
240 		break;							\
241 	case 8:								\
242 		put_unaligned_be64((__force u64)(val), __gu_p);		\
243 		break;							\
244 	default:							\
245 		__bad_unaligned_access_size();				\
246 		break;							\
247 	}								\
248 	(void)0; })
249 
250 #ifdef __LITTLE_ENDIAN
251 # define get_unaligned __get_unaligned_le
252 # define put_unaligned __put_unaligned_le
253 #else
254 # define get_unaligned __get_unaligned_be
255 # define put_unaligned __put_unaligned_be
256 #endif
257 
258 #endif /* __ASM_SH_UNALIGNED_SH4A_H */
259