xref: /optee_os/lib/libutils/ext/include/compiler.h (revision a94eb7ef1b31876775a77f262352d07a834b9fe3)
1 /* SPDX-License-Identifier: BSD-2-Clause */
2 /*
3  * Copyright (c) 2014, STMicroelectronics International N.V.
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions are met:
8  *
9  * 1. Redistributions of source code must retain the above copyright notice,
10  * this list of conditions and the following disclaimer.
11  *
12  * 2. Redistributions in binary form must reproduce the above copyright notice,
13  * this list of conditions and the following disclaimer in the documentation
14  * and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
17  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
20  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #ifndef COMPILER_H
30 #define COMPILER_H
31 
32 /*
33  * Macros that should be used instead of using __attribute__ directly to
34  * ease portability and make the code easier to read.
35  *
36  * Some of the defines below is known to sometimes cause conflicts when
37  * this file is included from xtest in normal world. It is assumed that
38  * the conflicting defines has the same meaning in that environment.
39  * Surrounding the troublesome defines with #ifndef should be enough.
40  */
41 #define __deprecated	__attribute__((deprecated))
42 #ifndef __packed
43 #define __packed	__attribute__((packed))
44 #endif
45 #define __weak		__attribute__((weak))
46 #ifndef __noreturn
47 #define __noreturn	__attribute__((noreturn))
48 #endif
49 #define __pure		__attribute__((pure))
50 #define __aligned(x)	__attribute__((aligned(x)))
51 #define __printf(a, b)	__attribute__((format(printf, a, b)))
52 #define __noinline	__attribute__((noinline))
53 #define __attr_const	__attribute__((__const__))
54 #ifndef __unused
55 #define __unused	__attribute__((unused))
56 #endif
57 #define __maybe_unused	__attribute__((unused))
58 #ifndef __used
59 #define __used		__attribute__((__used__))
60 #endif
61 #define __must_check	__attribute__((warn_unused_result))
62 #define __cold		__attribute__((__cold__))
63 #define __section(x)	__attribute__((section(x)))
64 #define __data		__section(".data")
65 #define __bss		__section(".bss")
66 #define __rodata	__section(".rodata")
67 #define __rodata_unpaged __section(".rodata.__unpaged")
68 #define __early_ta	__section(".rodata.early_ta")
69 #define __noprof	__attribute__((no_instrument_function))
70 
71 #define __compiler_bswap64(x)	__builtin_bswap64((x))
72 #define __compiler_bswap32(x)	__builtin_bswap32((x))
73 #define __compiler_bswap16(x)	__builtin_bswap16((x))
74 
75 #define __GCC_VERSION (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + \
76 		       __GNUC_PATCHLEVEL__)
77 
78 #if __GCC_VERSION >= 50100 && !defined(__CHECKER__)
79 #define __HAVE_BUILTIN_OVERFLOW 1
80 #endif
81 
82 #ifdef __HAVE_BUILTIN_OVERFLOW
83 #define __compiler_add_overflow(a, b, res) \
84 	__builtin_add_overflow((a), (b), (res))
85 
86 #define __compiler_sub_overflow(a, b, res) \
87 	__builtin_sub_overflow((a), (b), (res))
88 
89 #define __compiler_mul_overflow(a, b, res) \
90 	__builtin_mul_overflow((a), (b), (res))
91 #else /*!__HAVE_BUILTIN_OVERFLOW*/
92 
93 /*
94  * Copied/inspired from https://www.fefe.de/intof.html
95  */
96 
97 #define __INTOF_ASSIGN(dest, src) (__extension__({ \
98 	typeof(src) __intof_x = (src); \
99 	typeof(dest) __intof_y = __intof_x; \
100 	(((uintmax_t)__intof_x == (uintmax_t)__intof_y) && \
101 	 ((__intof_x < 1) == (__intof_y < 1)) ? \
102 		(void)((dest) = __intof_y) , 0 : 1); \
103 }))
104 
105 #define __INTOF_ADD(c, a, b) (__extension__({ \
106 	typeof(a) __intofa_a = (a); \
107 	typeof(b) __intofa_b = (b); \
108 	intmax_t __intofa_a_signed = __intofa_a; \
109 	uintmax_t __intofa_a_unsigned = __intofa_a; \
110 	intmax_t __intofa_b_signed = __intofa_b; \
111 	uintmax_t __intofa_b_unsigned = __intofa_b; \
112 	\
113 	__intofa_b < 1 ? \
114 		__intofa_a < 1 ? \
115 			((INTMAX_MIN - __intofa_b_signed <= \
116 			  __intofa_a_signed)) ? \
117 				__INTOF_ASSIGN((c), __intofa_a_signed + \
118 						    __intofa_b_signed) : 1 \
119 		: \
120 			((__intofa_a_unsigned >= (uintmax_t)-__intofa_b) ? \
121 				__INTOF_ASSIGN((c), __intofa_a_unsigned + \
122 						    __intofa_b_signed) \
123 			: \
124 				__INTOF_ASSIGN((c), \
125 					(intmax_t)(__intofa_a_unsigned + \
126 						   __intofa_b_signed))) \
127 	: \
128 		__intofa_a < 1 ? \
129 			((__intofa_b_unsigned >= (uintmax_t)-__intofa_a) ? \
130 				__INTOF_ASSIGN((c), __intofa_a_signed + \
131 						    __intofa_b_unsigned) \
132 			: \
133 				__INTOF_ASSIGN((c), \
134 					(intmax_t)(__intofa_a_signed + \
135 						   __intofa_b_unsigned))) \
136 		: \
137 			((UINTMAX_MAX - __intofa_b_unsigned >= \
138 			  __intofa_a_unsigned) ? \
139 				__INTOF_ASSIGN((c), __intofa_a_unsigned + \
140 						    __intofa_b_unsigned) : 1); \
141 }))
142 
143 #define __INTOF_SUB(c, a, b) (__extension__({ \
144 	typeof(a) __intofs_a = a; \
145 	typeof(b) __intofs_b = b; \
146 	intmax_t __intofs_a_signed = __intofs_a; \
147 	uintmax_t __intofs_a_unsigned = __intofs_a; \
148 	intmax_t __intofs_b_signed = __intofs_b; \
149 	uintmax_t __intofs_b_unsigned = __intofs_b; \
150 	\
151 	__intofs_b < 1 ? \
152 		__intofs_a < 1 ? \
153 			((INTMAX_MAX + __intofs_b >= __intofs_a) ? \
154 				__INTOF_ASSIGN((c), __intofs_a_signed - \
155 						    __intofs_b_signed) : 1) \
156 		: \
157 			(((uintmax_t)(UINTMAX_MAX + __intofs_b_signed) >= \
158 			  __intofs_a_unsigned) ? \
159 				__INTOF_ASSIGN((c), __intofs_a - \
160 						    __intofs_b) : 1) \
161 	: \
162 		__intofs_a < 1 ? \
163 			(((INTMAX_MIN + __intofs_b <= __intofs_a)) ? \
164 				__INTOF_ASSIGN((c), \
165 					(intmax_t)(__intofs_a_signed - \
166 						   __intofs_b_unsigned)) : 1) \
167 		: \
168 			((__intofs_b_unsigned <= __intofs_a_unsigned) ? \
169 				__INTOF_ASSIGN((c), __intofs_a_unsigned - \
170 						    __intofs_b_unsigned) \
171 			: \
172 				__INTOF_ASSIGN((c), \
173 					(intmax_t)(__intofs_a_unsigned - \
174 						   __intofs_b_unsigned))); \
175 }))
176 
177 /*
178  * Dealing with detecting overflow in multiplication of integers.
179  *
180  * First step is to remove two corner cases with the minum signed integer
181  * which can't be represented as a positive integer + sign.
182  * Multiply with 0 or 1 can't overflow, no checking needed of the operation,
183  * only if it can be assigned to the result.
184  *
185  * After the corner cases are eliminated we convert the two factors to
186  * positive unsigned values, keeping track of the original in another
187  * variable which is used at the end to determine the sign of the product.
188  *
189  * The two terms (a and b) are divided into upper and lower half (x1 upper
190  * and x0 lower), so the product is:
191  * ((a1 << hshift) + a0) * ((b1 << hshift) + b0)
192  * which also is:
193  * ((a1 * b1) << (hshift * 2)) +				(T1)
194  * ((a1 * b0 + a0 * b1) << hshift) +				(T2)
195  * (a0 * b0)							(T3)
196  *
197  * From this we can tell and (a1 * b1) has to be 0 or we'll overflow, that
198  * is, at least one of a1 or b1 has to be 0. Once this has been checked the
199  * addition: ((a1 * b0) << hshift) + ((a0 * b1) << hshift)
200  * isn't an addition as one of the terms will be 0.
201  *
202  * Since each factor in: (a0 * b0)
203  * only uses half the capicity of the underlaying type it can't overflow
204  *
205  * The addition of T2 and T3 can overflow so we use __INTOF_ADD() to
206  * perform that addition. If the addition succeeds without overflow the
207  * result is assigned the required sign and checked for overflow again.
208  */
209 
210 #define __intof_mul_negate	((__intof_oa < 1) != (__intof_ob < 1))
211 #define __intof_mul_hshift	(sizeof(uintmax_t) * 8 / 2)
212 #define __intof_mul_hmask	(UINTMAX_MAX >> __intof_mul_hshift)
213 #define __intof_mul_a0		((uintmax_t)(__intof_a) >> __intof_mul_hshift)
214 #define __intof_mul_b0		((uintmax_t)(__intof_b) >> __intof_mul_hshift)
215 #define __intof_mul_a1		((uintmax_t)(__intof_a) & __intof_mul_hmask)
216 #define __intof_mul_b1		((uintmax_t)(__intof_b) & __intof_mul_hmask)
217 #define __intof_mul_t		(__intof_mul_a1 * __intof_mul_b0 + \
218 				 __intof_mul_a0 * __intof_mul_b1)
219 
220 #define __INTOF_MUL(c, a, b) (__extension__({ \
221 	typeof(a) __intof_oa = (a); \
222 	typeof(a) __intof_a = __intof_oa < 1 ? -__intof_oa : __intof_oa; \
223 	typeof(b) __intof_ob = (b); \
224 	typeof(b) __intof_b = __intof_ob < 1 ? -__intof_ob : __intof_ob; \
225 	typeof(c) __intof_c; \
226 	\
227 	__intof_oa == 0 || __intof_ob == 0 || \
228 	__intof_oa == 1 || __intof_ob == 1 ? \
229 		__INTOF_ASSIGN((c), __intof_oa * __intof_ob) : \
230 	(__intof_mul_a0 && __intof_mul_b0) || \
231 	 __intof_mul_t > __intof_mul_hmask ?  1 : \
232 	__INTOF_ADD((__intof_c), __intof_mul_t << __intof_mul_hshift, \
233 				 __intof_mul_a1 * __intof_mul_b1) ? 1 : \
234 	__intof_mul_negate ? __INTOF_ASSIGN((c), -__intof_c) : \
235 			     __INTOF_ASSIGN((c), __intof_c); \
236 }))
237 
238 #define __compiler_add_overflow(a, b, res) __INTOF_ADD(*(res), (a), (b))
239 #define __compiler_sub_overflow(a, b, res) __INTOF_SUB(*(res), (a), (b))
240 #define __compiler_mul_overflow(a, b, res) __INTOF_MUL(*(res), (a), (b))
241 
242 #endif /*!__HAVE_BUILTIN_OVERFLOW*/
243 
244 #define __compiler_compare_and_swap(p, oval, nval) \
245 	__atomic_compare_exchange_n((p), (oval), (nval), true, \
246 				    __ATOMIC_ACQUIRE, __ATOMIC_RELAXED) \
247 
248 #define __compiler_atomic_load(p) __atomic_load_n((p), __ATOMIC_RELAXED)
249 #define __compiler_atomic_store(p, val) \
250 	__atomic_store_n((p), (val), __ATOMIC_RELAXED)
251 
252 #endif /*COMPILER_H*/
253