xref: /optee_os/lib/libutils/ext/include/compiler.h (revision c44d734b6366cbf4d12610310e809872db65f89d)
1 /* SPDX-License-Identifier: BSD-2-Clause */
2 /*
3  * Copyright (c) 2014, STMicroelectronics International N.V.
4  */
5 
6 #ifndef COMPILER_H
7 #define COMPILER_H
8 
9 /*
10  * Macros that should be used instead of using __attribute__ directly to
11  * ease portability and make the code easier to read.
12  *
13  * Some of the defines below is known to sometimes cause conflicts when
14  * this file is included from xtest in normal world. It is assumed that
15  * the conflicting defines has the same meaning in that environment.
16  * Surrounding the troublesome defines with #ifndef should be enough.
17  */
18 #define __deprecated	__attribute__((deprecated))
19 #ifndef __packed
20 #define __packed	__attribute__((packed))
21 #endif
22 #define __weak		__attribute__((weak))
23 #ifndef __noreturn
24 #define __noreturn	__attribute__((__noreturn__))
25 #endif
26 #define __pure		__attribute__((pure))
27 #define __aligned(x)	__attribute__((aligned(x)))
28 #define __printf(a, b)	__attribute__((format(printf, a, b)))
29 #define __noinline	__attribute__((noinline))
30 #define __attr_const	__attribute__((__const__))
31 #ifndef __unused
32 #define __unused	__attribute__((unused))
33 #endif
34 #define __maybe_unused	__attribute__((unused))
35 #ifndef __used
36 #define __used		__attribute__((__used__))
37 #endif
38 #define __must_check	__attribute__((warn_unused_result))
39 #define __cold		__attribute__((__cold__))
40 #define __section(x)	__attribute__((section(x)))
41 #define __data		__section(".data")
42 #define __bss		__section(".bss")
43 #ifdef __clang__
44 #define __SECTION_FLAGS_RODATA
45 #else
46 /*
47  * Override sections flags/type generated by the C compiler to make sure they
48  * are: "a",%progbits (thus creating an allocatable, non-writeable, non-
49  * executable data section).
50  * The trailing COMMENT_CHAR comments out the flags generated by the compiler.
51  * This avoids a harmless warning with GCC.
52  */
53 #if defined(__aarch64__) || defined(__arm__)
54 #define COMMENT_CHAR "//"
55 #else
56 #define COMMENT_CHAR "#"
57 #endif
58 #define __SECTION_FLAGS_RODATA ",\"a\",%progbits " COMMENT_CHAR
59 #endif
60 #define __rodata	__section(".rodata" __SECTION_FLAGS_RODATA)
61 #define __rodata_dummy	__section(".rodata.dummy" __SECTION_FLAGS_RODATA)
62 #define __rodata_unpaged(x) \
63 	__section(".rodata.__unpaged." x __SECTION_FLAGS_RODATA)
64 #ifdef CFG_CORE_ASLR
65 #define __relrodata_unpaged(x) __section(".data.rel.ro.__unpaged." x)
66 #else
67 #define __relrodata_unpaged(x) __rodata_unpaged(x)
68 #endif
69 #ifdef CFG_VIRTUALIZATION
70 #define __nex_bss		__section(".nex_bss")
71 #define __nex_data		__section(".nex_data")
72 #else  /* CFG_VIRTUALIZATION */
73 #define __nex_bss
74 #define __nex_data
75 #endif	/* CFG_VIRTUALIZATION */
76 #define __noprof	__attribute__((no_instrument_function))
77 #define __nostackcheck	__attribute__((no_instrument_function))
78 
79 #define __compiler_bswap64(x)	__builtin_bswap64((x))
80 #define __compiler_bswap32(x)	__builtin_bswap32((x))
81 #define __compiler_bswap16(x)	__builtin_bswap16((x))
82 
83 #define __GCC_VERSION (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + \
84 		       __GNUC_PATCHLEVEL__)
85 
86 #if __GCC_VERSION >= 50100 && !defined(__CHECKER__)
87 #define __HAVE_BUILTIN_OVERFLOW 1
88 #endif
89 
90 #if __GCC_VERSION >= 90100 && !defined(__CHECKER__)
91 #define __HAVE_SINGLE_ARGUMENT_STATIC_ASSERT 1
92 #endif
93 
94 #ifdef __HAVE_BUILTIN_OVERFLOW
95 #define __compiler_add_overflow(a, b, res) \
96 	__builtin_add_overflow((a), (b), (res))
97 
98 #define __compiler_sub_overflow(a, b, res) \
99 	__builtin_sub_overflow((a), (b), (res))
100 
101 #define __compiler_mul_overflow(a, b, res) \
102 	__builtin_mul_overflow((a), (b), (res))
103 #else /*!__HAVE_BUILTIN_OVERFLOW*/
104 
105 /*
106  * Copied/inspired from https://www.fefe.de/intof.html
107  */
108 
109 #define __INTOF_ASSIGN(dest, src) (__extension__({ \
110 	typeof(src) __intof_x = (src); \
111 	typeof(dest) __intof_y = __intof_x; \
112 	(((uintmax_t)__intof_x == (uintmax_t)__intof_y) && \
113 	 ((__intof_x < 1) == (__intof_y < 1)) ? \
114 		(void)((dest) = __intof_y) , 0 : 1); \
115 }))
116 
117 #define __INTOF_ADD(c, a, b) (__extension__({ \
118 	typeof(a) __intofa_a = (a); \
119 	typeof(b) __intofa_b = (b); \
120 	intmax_t __intofa_a_signed = __intofa_a; \
121 	uintmax_t __intofa_a_unsigned = __intofa_a; \
122 	intmax_t __intofa_b_signed = __intofa_b; \
123 	uintmax_t __intofa_b_unsigned = __intofa_b; \
124 	\
125 	__intofa_b < 1 ? \
126 		__intofa_a < 1 ? \
127 			((INTMAX_MIN - __intofa_b_signed <= \
128 			  __intofa_a_signed)) ? \
129 				__INTOF_ASSIGN((c), __intofa_a_signed + \
130 						    __intofa_b_signed) : 1 \
131 		: \
132 			((__intofa_a_unsigned >= (uintmax_t)-__intofa_b) ? \
133 				__INTOF_ASSIGN((c), __intofa_a_unsigned + \
134 						    __intofa_b_signed) \
135 			: \
136 				__INTOF_ASSIGN((c), \
137 					(intmax_t)(__intofa_a_unsigned + \
138 						   __intofa_b_signed))) \
139 	: \
140 		__intofa_a < 1 ? \
141 			((__intofa_b_unsigned >= (uintmax_t)-__intofa_a) ? \
142 				__INTOF_ASSIGN((c), __intofa_a_signed + \
143 						    __intofa_b_unsigned) \
144 			: \
145 				__INTOF_ASSIGN((c), \
146 					(intmax_t)(__intofa_a_signed + \
147 						   __intofa_b_unsigned))) \
148 		: \
149 			((UINTMAX_MAX - __intofa_b_unsigned >= \
150 			  __intofa_a_unsigned) ? \
151 				__INTOF_ASSIGN((c), __intofa_a_unsigned + \
152 						    __intofa_b_unsigned) : 1); \
153 }))
154 
155 #define __INTOF_SUB(c, a, b) (__extension__({ \
156 	typeof(a) __intofs_a = a; \
157 	typeof(b) __intofs_b = b; \
158 	intmax_t __intofs_a_signed = __intofs_a; \
159 	uintmax_t __intofs_a_unsigned = __intofs_a; \
160 	intmax_t __intofs_b_signed = __intofs_b; \
161 	uintmax_t __intofs_b_unsigned = __intofs_b; \
162 	\
163 	__intofs_b < 1 ? \
164 		__intofs_a < 1 ? \
165 			((INTMAX_MAX + __intofs_b_signed >= \
166 			  __intofs_a_signed) ? \
167 				__INTOF_ASSIGN((c), __intofs_a_signed - \
168 						    __intofs_b_signed) : 1) \
169 		: \
170 			(((uintmax_t)(UINTMAX_MAX + __intofs_b_signed) >= \
171 			  __intofs_a_unsigned) ? \
172 				__INTOF_ASSIGN((c), __intofs_a - \
173 						    __intofs_b) : 1) \
174 	: \
175 		__intofs_a < 1 ? \
176 			(((intmax_t)(INTMAX_MIN + __intofs_b) <= \
177 			  __intofs_a_signed) ? \
178 				__INTOF_ASSIGN((c), \
179 					(intmax_t)(__intofs_a_signed - \
180 						   __intofs_b_unsigned)) : 1) \
181 		: \
182 			((__intofs_b_unsigned <= __intofs_a_unsigned) ? \
183 				__INTOF_ASSIGN((c), __intofs_a_unsigned - \
184 						    __intofs_b_unsigned) \
185 			: \
186 				__INTOF_ASSIGN((c), \
187 					(intmax_t)(__intofs_a_unsigned - \
188 						   __intofs_b_unsigned))); \
189 }))
190 
191 /*
192  * Dealing with detecting overflow in multiplication of integers.
193  *
194  * First step is to remove two corner cases with the minum signed integer
195  * which can't be represented as a positive integer + sign.
196  * Multiply with 0 or 1 can't overflow, no checking needed of the operation,
197  * only if it can be assigned to the result.
198  *
199  * After the corner cases are eliminated we convert the two factors to
200  * positive unsigned values, keeping track of the original in another
201  * variable which is used at the end to determine the sign of the product.
202  *
203  * The two terms (a and b) are divided into upper and lower half (x1 upper
204  * and x0 lower), so the product is:
205  * ((a1 << hshift) + a0) * ((b1 << hshift) + b0)
206  * which also is:
207  * ((a1 * b1) << (hshift * 2)) +				(T1)
208  * ((a1 * b0 + a0 * b1) << hshift) +				(T2)
209  * (a0 * b0)							(T3)
210  *
211  * From this we can tell and (a1 * b1) has to be 0 or we'll overflow, that
212  * is, at least one of a1 or b1 has to be 0. Once this has been checked the
213  * addition: ((a1 * b0) << hshift) + ((a0 * b1) << hshift)
214  * isn't an addition as one of the terms will be 0.
215  *
216  * Since each factor in: (a0 * b0)
217  * only uses half the capicity of the underlaying type it can't overflow
218  *
219  * The addition of T2 and T3 can overflow so we use __INTOF_ADD() to
220  * perform that addition. If the addition succeeds without overflow the
221  * result is assigned the required sign and checked for overflow again.
222  */
223 
224 #define __intof_mul_negate	((__intof_oa < 1) != (__intof_ob < 1))
225 #define __intof_mul_hshift	(sizeof(uintmax_t) * 8 / 2)
226 #define __intof_mul_hmask	(UINTMAX_MAX >> __intof_mul_hshift)
227 #define __intof_mul_a0		((uintmax_t)(__intof_a) >> __intof_mul_hshift)
228 #define __intof_mul_b0		((uintmax_t)(__intof_b) >> __intof_mul_hshift)
229 #define __intof_mul_a1		((uintmax_t)(__intof_a) & __intof_mul_hmask)
230 #define __intof_mul_b1		((uintmax_t)(__intof_b) & __intof_mul_hmask)
231 #define __intof_mul_t		(__intof_mul_a1 * __intof_mul_b0 + \
232 				 __intof_mul_a0 * __intof_mul_b1)
233 
234 #define __INTOF_MUL(c, a, b) (__extension__({ \
235 	typeof(a) __intof_oa = (a); \
236 	typeof(a) __intof_a = __intof_oa < 1 ? -__intof_oa : __intof_oa; \
237 	typeof(b) __intof_ob = (b); \
238 	typeof(b) __intof_b = __intof_ob < 1 ? -__intof_ob : __intof_ob; \
239 	typeof(c) __intof_c; \
240 	\
241 	__intof_oa == 0 || __intof_ob == 0 || \
242 	__intof_oa == 1 || __intof_ob == 1 ? \
243 		__INTOF_ASSIGN((c), __intof_oa * __intof_ob) : \
244 	(__intof_mul_a0 && __intof_mul_b0) || \
245 	 __intof_mul_t > __intof_mul_hmask ?  1 : \
246 	__INTOF_ADD((__intof_c), __intof_mul_t << __intof_mul_hshift, \
247 				 __intof_mul_a1 * __intof_mul_b1) ? 1 : \
248 	__intof_mul_negate ? __INTOF_ASSIGN((c), -__intof_c) : \
249 			     __INTOF_ASSIGN((c), __intof_c); \
250 }))
251 
252 #define __compiler_add_overflow(a, b, res) __INTOF_ADD(*(res), (a), (b))
253 #define __compiler_sub_overflow(a, b, res) __INTOF_SUB(*(res), (a), (b))
254 #define __compiler_mul_overflow(a, b, res) __INTOF_MUL(*(res), (a), (b))
255 
256 #endif /*!__HAVE_BUILTIN_OVERFLOW*/
257 
258 #define __compiler_compare_and_swap(p, oval, nval) \
259 	__atomic_compare_exchange_n((p), (oval), (nval), true, \
260 				    __ATOMIC_ACQUIRE, __ATOMIC_RELAXED) \
261 
262 #define __compiler_atomic_load(p) __atomic_load_n((p), __ATOMIC_RELAXED)
263 #define __compiler_atomic_store(p, val) \
264 	__atomic_store_n((p), (val), __ATOMIC_RELAXED)
265 
266 #define barrier() asm volatile ("" : : : "memory")
267 
268 #ifndef __has_attribute
269 #define __has_attribute(x) 0
270 #endif
271 
272 #if __has_attribute(__fallthrough__)
273 #define fallthrough __attribute__((__fallthrough__))
274 #else
275 #define fallthrough do {} while (0) /* fallthrough */
276 #endif
277 
278 #endif /*COMPILER_H*/
279