xref: /optee_os/lib/libutils/ext/include/compiler.h (revision e3fb2bd005f0b06754cc2181b881e62a05424c35)
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 #define __alias(x)	__attribute__((alias(x)))
24 #ifndef __noreturn
25 #define __noreturn	__attribute__((__noreturn__))
26 #endif
27 #ifndef __no_stack_protector
28 #define __no_stack_protector __attribute__((no_stack_protector))
29 #endif
30 #define __pure		__attribute__((pure))
31 #define __aligned(x)	__attribute__((aligned(x)))
32 #define __printf(a, b)	__attribute__((format(printf, a, b)))
33 #define __noinline	__attribute__((noinline))
34 #define __attr_const	__attribute__((__const__))
35 #ifndef __unused
36 #define __unused	__attribute__((unused))
37 #endif
38 #define __maybe_unused	__attribute__((unused))
39 #ifndef __used
40 #define __used		__attribute__((__used__))
41 #endif
42 #define __must_check	__attribute__((warn_unused_result))
43 #define __cold		__attribute__((__cold__))
44 #define __section(x)	__attribute__((section(x)))
45 #define __data		__section(".data")
46 #define __bss		__section(".bss")
47 #ifdef __clang__
48 #define __SECTION_FLAGS_RODATA
49 #else
50 /*
51  * Override sections flags/type generated by the C compiler to make sure they
52  * are: "a",%progbits (thus creating an allocatable, non-writeable, non-
53  * executable data section).
54  * The trailing COMMENT_CHAR comments out the flags generated by the compiler.
55  * This avoids a harmless warning with GCC.
56  */
57 #if defined(__aarch64__) || defined(__arm__)
58 #define COMMENT_CHAR "//"
59 #else
60 #define COMMENT_CHAR "#"
61 #endif
62 #define __SECTION_FLAGS_RODATA ",\"a\",%progbits " COMMENT_CHAR
63 #endif
64 #define __rodata	__section(".rodata" __SECTION_FLAGS_RODATA)
65 #define __rodata_dummy	__section(".rodata.dummy" __SECTION_FLAGS_RODATA)
66 #define __rodata_unpaged(x) \
67 	__section(".rodata.__unpaged." x __SECTION_FLAGS_RODATA)
68 #ifdef CFG_CORE_ASLR
69 #define __relrodata_unpaged(x) __section(".data.rel.ro.__unpaged." x)
70 #else
71 #define __relrodata_unpaged(x) __rodata_unpaged(x)
72 #endif
73 #ifdef CFG_NS_VIRTUALIZATION
74 #define __nex_bss		__section(".nex_bss")
75 #define __nex_data		__section(".nex_data")
76 #else  /* CFG_NS_VIRTUALIZATION */
77 #define __nex_bss
78 #define __nex_data
79 #endif	/* CFG_NS_VIRTUALIZATION */
80 #define __noprof	__attribute__((no_instrument_function))
81 #define __nostackcheck	__attribute__((no_instrument_function))
82 
83 #define __compiler_bswap64(x)	__builtin_bswap64((x))
84 #define __compiler_bswap32(x)	__builtin_bswap32((x))
85 #define __compiler_bswap16(x)	__builtin_bswap16((x))
86 
87 #define __GCC_VERSION (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + \
88 		       __GNUC_PATCHLEVEL__)
89 
90 #if __GCC_VERSION >= 50100 && !defined(__CHECKER__)
91 #define __HAVE_BUILTIN_OVERFLOW 1
92 #endif
93 
94 #if __GCC_VERSION >= 90100 && !defined(__CHECKER__)
95 #define __HAVE_SINGLE_ARGUMENT_STATIC_ASSERT 1
96 #endif
97 
98 #ifdef __HAVE_BUILTIN_OVERFLOW
99 #define __compiler_add_overflow(a, b, res) \
100 	__builtin_add_overflow((a), (b), (res))
101 
102 #define __compiler_sub_overflow(a, b, res) \
103 	__builtin_sub_overflow((a), (b), (res))
104 
105 #define __compiler_mul_overflow(a, b, res) \
106 	__builtin_mul_overflow((a), (b), (res))
107 #else /*!__HAVE_BUILTIN_OVERFLOW*/
108 
109 /*
110  * Copied/inspired from https://www.fefe.de/intof.html
111  */
112 
113 #define __INTOF_ASSIGN(dest, src) (__extension__({ \
114 	typeof(src) __intof_x = (src); \
115 	typeof(dest) __intof_y = __intof_x; \
116 	(((uintmax_t)__intof_x == (uintmax_t)__intof_y) && \
117 	 ((__intof_x < 1) == (__intof_y < 1)) ? \
118 		(void)((dest) = __intof_y) , 0 : 1); \
119 }))
120 
121 #define __INTOF_ADD(c, a, b) (__extension__({ \
122 	typeof(a) __intofa_a = (a); \
123 	typeof(b) __intofa_b = (b); \
124 	intmax_t __intofa_a_signed = __intofa_a; \
125 	uintmax_t __intofa_a_unsigned = __intofa_a; \
126 	intmax_t __intofa_b_signed = __intofa_b; \
127 	uintmax_t __intofa_b_unsigned = __intofa_b; \
128 	\
129 	__intofa_b < 1 ? \
130 		__intofa_a < 1 ? \
131 			((INTMAX_MIN - __intofa_b_signed <= \
132 			  __intofa_a_signed)) ? \
133 				__INTOF_ASSIGN((c), __intofa_a_signed + \
134 						    __intofa_b_signed) : 1 \
135 		: \
136 			((__intofa_a_unsigned >= (uintmax_t)-__intofa_b) ? \
137 				__INTOF_ASSIGN((c), __intofa_a_unsigned + \
138 						    __intofa_b_signed) \
139 			: \
140 				__INTOF_ASSIGN((c), \
141 					(intmax_t)(__intofa_a_unsigned + \
142 						   __intofa_b_signed))) \
143 	: \
144 		__intofa_a < 1 ? \
145 			((__intofa_b_unsigned >= (uintmax_t)-__intofa_a) ? \
146 				__INTOF_ASSIGN((c), __intofa_a_signed + \
147 						    __intofa_b_unsigned) \
148 			: \
149 				__INTOF_ASSIGN((c), \
150 					(intmax_t)(__intofa_a_signed + \
151 						   __intofa_b_unsigned))) \
152 		: \
153 			((UINTMAX_MAX - __intofa_b_unsigned >= \
154 			  __intofa_a_unsigned) ? \
155 				__INTOF_ASSIGN((c), __intofa_a_unsigned + \
156 						    __intofa_b_unsigned) : 1); \
157 }))
158 
159 #define __INTOF_SUB(c, a, b) (__extension__({ \
160 	typeof(a) __intofs_a = a; \
161 	typeof(b) __intofs_b = b; \
162 	intmax_t __intofs_a_signed = __intofs_a; \
163 	uintmax_t __intofs_a_unsigned = __intofs_a; \
164 	intmax_t __intofs_b_signed = __intofs_b; \
165 	uintmax_t __intofs_b_unsigned = __intofs_b; \
166 	\
167 	__intofs_b < 1 ? \
168 		__intofs_a < 1 ? \
169 			((INTMAX_MAX + __intofs_b_signed >= \
170 			  __intofs_a_signed) ? \
171 				__INTOF_ASSIGN((c), __intofs_a_signed - \
172 						    __intofs_b_signed) : 1) \
173 		: \
174 			(((uintmax_t)(UINTMAX_MAX + __intofs_b_signed) >= \
175 			  __intofs_a_unsigned) ? \
176 				__INTOF_ASSIGN((c), __intofs_a - \
177 						    __intofs_b) : 1) \
178 	: \
179 		__intofs_a < 1 ? \
180 			(((intmax_t)(INTMAX_MIN + __intofs_b) <= \
181 			  __intofs_a_signed) ? \
182 				__INTOF_ASSIGN((c), \
183 					(intmax_t)(__intofs_a_signed - \
184 						   __intofs_b_unsigned)) : 1) \
185 		: \
186 			((__intofs_b_unsigned <= __intofs_a_unsigned) ? \
187 				__INTOF_ASSIGN((c), __intofs_a_unsigned - \
188 						    __intofs_b_unsigned) \
189 			: \
190 				__INTOF_ASSIGN((c), \
191 					(intmax_t)(__intofs_a_unsigned - \
192 						   __intofs_b_unsigned))); \
193 }))
194 
195 /*
196  * Dealing with detecting overflow in multiplication of integers.
197  *
198  * First step is to remove two corner cases with the minum signed integer
199  * which can't be represented as a positive integer + sign.
200  * Multiply with 0 or 1 can't overflow, no checking needed of the operation,
201  * only if it can be assigned to the result.
202  *
203  * After the corner cases are eliminated we convert the two factors to
204  * positive unsigned values, keeping track of the original in another
205  * variable which is used at the end to determine the sign of the product.
206  *
207  * The two terms (a and b) are divided into upper and lower half (x1 upper
208  * and x0 lower), so the product is:
209  * ((a1 << hshift) + a0) * ((b1 << hshift) + b0)
210  * which also is:
211  * ((a1 * b1) << (hshift * 2)) +				(T1)
212  * ((a1 * b0 + a0 * b1) << hshift) +				(T2)
213  * (a0 * b0)							(T3)
214  *
215  * From this we can tell and (a1 * b1) has to be 0 or we'll overflow, that
216  * is, at least one of a1 or b1 has to be 0. Once this has been checked the
217  * addition: ((a1 * b0) << hshift) + ((a0 * b1) << hshift)
218  * isn't an addition as one of the terms will be 0.
219  *
220  * Since each factor in: (a0 * b0)
221  * only uses half the capicity of the underlaying type it can't overflow
222  *
223  * The addition of T2 and T3 can overflow so we use __INTOF_ADD() to
224  * perform that addition. If the addition succeeds without overflow the
225  * result is assigned the required sign and checked for overflow again.
226  */
227 
228 #define __intof_mul_negate	((__intof_oa < 1) != (__intof_ob < 1))
229 #define __intof_mul_hshift	(sizeof(uintmax_t) * 8 / 2)
230 #define __intof_mul_hmask	(UINTMAX_MAX >> __intof_mul_hshift)
231 #define __intof_mul_a0		((uintmax_t)(__intof_a) >> __intof_mul_hshift)
232 #define __intof_mul_b0		((uintmax_t)(__intof_b) >> __intof_mul_hshift)
233 #define __intof_mul_a1		((uintmax_t)(__intof_a) & __intof_mul_hmask)
234 #define __intof_mul_b1		((uintmax_t)(__intof_b) & __intof_mul_hmask)
235 #define __intof_mul_t		(__intof_mul_a1 * __intof_mul_b0 + \
236 				 __intof_mul_a0 * __intof_mul_b1)
237 
238 #define __INTOF_MUL(c, a, b) (__extension__({ \
239 	typeof(a) __intof_oa = (a); \
240 	typeof(a) __intof_a = __intof_oa < 1 ? -__intof_oa : __intof_oa; \
241 	typeof(b) __intof_ob = (b); \
242 	typeof(b) __intof_b = __intof_ob < 1 ? -__intof_ob : __intof_ob; \
243 	typeof(c) __intof_c; \
244 	\
245 	__intof_oa == 0 || __intof_ob == 0 || \
246 	__intof_oa == 1 || __intof_ob == 1 ? \
247 		__INTOF_ASSIGN((c), __intof_oa * __intof_ob) : \
248 	(__intof_mul_a0 && __intof_mul_b0) || \
249 	 __intof_mul_t > __intof_mul_hmask ?  1 : \
250 	__INTOF_ADD((__intof_c), __intof_mul_t << __intof_mul_hshift, \
251 				 __intof_mul_a1 * __intof_mul_b1) ? 1 : \
252 	__intof_mul_negate ? __INTOF_ASSIGN((c), -__intof_c) : \
253 			     __INTOF_ASSIGN((c), __intof_c); \
254 }))
255 
256 #define __compiler_add_overflow(a, b, res) __INTOF_ADD(*(res), (a), (b))
257 #define __compiler_sub_overflow(a, b, res) __INTOF_SUB(*(res), (a), (b))
258 #define __compiler_mul_overflow(a, b, res) __INTOF_MUL(*(res), (a), (b))
259 
260 #endif /*!__HAVE_BUILTIN_OVERFLOW*/
261 
262 #define __compiler_compare_and_swap(p, oval, nval) \
263 	__atomic_compare_exchange_n((p), (oval), (nval), true, \
264 				    __ATOMIC_ACQUIRE, __ATOMIC_RELAXED) \
265 
266 #define __compiler_atomic_load(p) __atomic_load_n((p), __ATOMIC_RELAXED)
267 #define __compiler_atomic_store(p, val) \
268 	__atomic_store_n((p), (val), __ATOMIC_RELAXED)
269 
270 #define barrier() asm volatile ("" : : : "memory")
271 
272 #ifndef __has_attribute
273 #define __has_attribute(x) 0
274 #endif
275 
276 #if __has_attribute(__fallthrough__)
277 #define fallthrough __attribute__((__fallthrough__))
278 #else
279 #define fallthrough do {} while (0) /* fallthrough */
280 #endif
281 
282 #ifndef __clang__
283 #define __no_stackprot __attribute__((__optimize__ ("-fno-stack-protector")))
284 #else
285 #define __no_stackprot
286 #endif
287 
288 #define __inhibit_loop_to_libcall \
289 	__attribute__ ((__optimize__ ("-fno-tree-loop-distribute-patterns")))
290 #endif /*COMPILER_H*/
291