xref: /optee_os/lib/libutils/ext/include/compiler.h (revision ecdedc94e72049d79afda94488b8d67974eb7583)
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 #define __INTOF_HALF_MAX_SIGNED(type) ((type)1 << (sizeof(type)*8-2))
97 #define __INTOF_MAX_SIGNED(type) (__INTOF_HALF_MAX_SIGNED(type) - 1 + \
98 			    __INTOF_HALF_MAX_SIGNED(type))
99 #define __INTOF_MIN_SIGNED(type) (-1 - __INTOF_MAX_SIGNED(type))
100 
101 #define __INTOF_MIN(type) ((type)-1 < 1?__INTOF_MIN_SIGNED(type):(type)0)
102 #define __INTOF_MAX(type) ((type)~__INTOF_MIN(type))
103 
104 #define __INTOF_ASSIGN(dest, src) (__extension__({ \
105 	typeof(src) __intof_x = (src); \
106 	typeof(dest) __intof_y = __intof_x; \
107 	(((uintmax_t)__intof_x == (uintmax_t)__intof_y) && \
108 	 ((__intof_x < 1) == (__intof_y < 1)) ? \
109 		(void)((dest) = __intof_y) , 0 : 1); \
110 }))
111 
112 #define __INTOF_ADD(c, a, b) (__extension__({ \
113 	typeof(a) __intofa_a = (a); \
114 	typeof(b) __intofa_b = (b); \
115 	intmax_t __intofa_a_signed = __intofa_a; \
116 	uintmax_t __intofa_a_unsigned = __intofa_a; \
117 	intmax_t __intofa_b_signed = __intofa_b; \
118 	uintmax_t __intofa_b_unsigned = __intofa_b; \
119 	\
120 	__intofa_b < 1 ? \
121 		__intofa_a < 1 ? \
122 			((INTMAX_MIN - __intofa_b_signed <= \
123 			  __intofa_a_signed)) ? \
124 				__INTOF_ASSIGN((c), __intofa_a_signed + \
125 						    __intofa_b_signed) : 1 \
126 		: \
127 			((__intofa_a_unsigned >= (uintmax_t)-__intofa_b) ? \
128 				__INTOF_ASSIGN((c), __intofa_a_unsigned + \
129 						    __intofa_b_signed) \
130 			: \
131 				__INTOF_ASSIGN((c), \
132 					(intmax_t)(__intofa_a_unsigned + \
133 						   __intofa_b_signed))) \
134 	: \
135 		__intofa_a < 1 ? \
136 			((__intofa_b_unsigned >= (uintmax_t)-__intofa_a) ? \
137 				__INTOF_ASSIGN((c), __intofa_a_signed + \
138 						    __intofa_b_unsigned) \
139 			: \
140 				__INTOF_ASSIGN((c), \
141 					(intmax_t)(__intofa_a_signed + \
142 						   __intofa_b_unsigned))) \
143 		: \
144 			((UINTMAX_MAX - __intofa_b_unsigned >= \
145 			  __intofa_a_unsigned) ? \
146 				__INTOF_ASSIGN((c), __intofa_a_unsigned + \
147 						    __intofa_b_unsigned) : 1); \
148 }))
149 
150 #define __INTOF_SUB(c, a, b) (__extension__({ \
151 	typeof(a) __intofs_a = a; \
152 	typeof(b) __intofs_b = b; \
153 	intmax_t __intofs_a_signed = __intofs_a; \
154 	uintmax_t __intofs_a_unsigned = __intofs_a; \
155 	intmax_t __intofs_b_signed = __intofs_b; \
156 	uintmax_t __intofs_b_unsigned = __intofs_b; \
157 	\
158 	__intofs_b < 1 ? \
159 		__intofs_a < 1 ? \
160 			((INTMAX_MAX + __intofs_b >= __intofs_a) ? \
161 				__INTOF_ASSIGN((c), __intofs_a_signed - \
162 						    __intofs_b_signed) : 1) \
163 		: \
164 			(((uintmax_t)(UINTMAX_MAX + __intofs_b_signed) >= \
165 			  __intofs_a_unsigned) ? \
166 				__INTOF_ASSIGN((c), __intofs_a - \
167 						    __intofs_b) : 1) \
168 	: \
169 		__intofs_a < 1 ? \
170 			(((INTMAX_MIN + __intofs_b <= __intofs_a)) ? \
171 				__INTOF_ASSIGN((c), \
172 					(intmax_t)(__intofs_a_signed - \
173 						   __intofs_b_unsigned)) : 1) \
174 		: \
175 			((__intofs_b_unsigned <= __intofs_a_unsigned) ? \
176 				__INTOF_ASSIGN((c), __intofs_a_unsigned - \
177 						    __intofs_b_unsigned) \
178 			: \
179 				__INTOF_ASSIGN((c), \
180 					(intmax_t)(__intofs_a_unsigned - \
181 						   __intofs_b_unsigned))); \
182 }))
183 
184 /*
185  * Dealing with detecting overflow in multiplication of integers.
186  *
187  * First step is to remove two corner cases with the minum signed integer
188  * which can't be represented as a positive integer + sign.
189  * Multiply with 0 or 1 can't overflow, no checking needed of the operation,
190  * only if it can be assigned to the result.
191  *
192  * After the corner cases are eliminated we convert the two factors to
193  * positive unsigned values, keeping track of the original in another
194  * variable which is used at the end to determine the sign of the product.
195  *
196  * The two terms (a and b) are divided into upper and lower half (x1 upper
197  * and x0 lower), so the product is:
198  * ((a1 << hshift) + a0) * ((b1 << hshift) + b0)
199  * which also is:
200  * ((a1 * b1) << (hshift * 2)) +				(T1)
201  * ((a1 * b0 + a0 * b1) << hshift) +				(T2)
202  * (a0 * b0)							(T3)
203  *
204  * From this we can tell and (a1 * b1) has to be 0 or we'll overflow, that
205  * is, at least one of a1 or b1 has to be 0. Once this has been checked the
206  * addition: ((a1 * b0) << hshift) + ((a0 * b1) << hshift)
207  * isn't an addition as one of the terms will be 0.
208  *
209  * Since each factor in: (a0 * b0)
210  * only uses half the capicity of the underlaying type it can't overflow
211  *
212  * The addition of T2 and T3 can overflow so we use __INTOF_ADD() to
213  * perform that addition. If the addition succeeds without overflow the
214  * result is assigned the required sign and checked for overflow again.
215  */
216 
217 #define __intof_mul_negate	((__intof_oa < 1) != (__intof_ob < 1))
218 #define __intof_mul_hshift	(sizeof(uintmax_t) * 8 / 2)
219 #define __intof_mul_hmask	(UINTMAX_MAX >> __intof_mul_hshift)
220 #define __intof_mul_a0		((uintmax_t)(__intof_a) >> __intof_mul_hshift)
221 #define __intof_mul_b0		((uintmax_t)(__intof_b) >> __intof_mul_hshift)
222 #define __intof_mul_a1		((uintmax_t)(__intof_a) & __intof_mul_hmask)
223 #define __intof_mul_b1		((uintmax_t)(__intof_b) & __intof_mul_hmask)
224 #define __intof_mul_t		(__intof_mul_a1 * __intof_mul_b0 + \
225 				 __intof_mul_a0 * __intof_mul_b1)
226 
227 #define __INTOF_MUL(c, a, b) (__extension__({ \
228 	typeof(a) __intof_oa = (a); \
229 	typeof(a) __intof_a = __intof_oa < 1 ? -__intof_oa : __intof_oa; \
230 	typeof(b) __intof_ob = (b); \
231 	typeof(b) __intof_b = __intof_ob < 1 ? -__intof_ob : __intof_ob; \
232 	typeof(c) __intof_c; \
233 	\
234 	__intof_oa == 0 || __intof_ob == 0 || \
235 	__intof_oa == 1 || __intof_ob == 1 ? \
236 		__INTOF_ASSIGN((c), __intof_oa * __intof_ob) : \
237 	(__intof_mul_a0 && __intof_mul_b0) || \
238 	 __intof_mul_t > __intof_mul_hmask ?  1 : \
239 	__INTOF_ADD((__intof_c), __intof_mul_t << __intof_mul_hshift, \
240 				 __intof_mul_a1 * __intof_mul_b1) ? 1 : \
241 	__intof_mul_negate ? __INTOF_ASSIGN((c), -__intof_c) : \
242 			     __INTOF_ASSIGN((c), __intof_c); \
243 }))
244 
245 #define __compiler_add_overflow(a, b, res) __INTOF_ADD(*(res), (a), (b))
246 #define __compiler_sub_overflow(a, b, res) __INTOF_SUB(*(res), (a), (b))
247 #define __compiler_mul_overflow(a, b, res) __INTOF_MUL(*(res), (a), (b))
248 
249 #endif /*!__HAVE_BUILTIN_OVERFLOW*/
250 
251 #define __compiler_compare_and_swap(p, oval, nval) \
252 	__atomic_compare_exchange_n((p), (oval), (nval), true, \
253 				    __ATOMIC_ACQUIRE, __ATOMIC_RELAXED) \
254 
255 #define __compiler_atomic_load(p) __atomic_load_n((p), __ATOMIC_RELAXED)
256 #define __compiler_atomic_store(p, val) \
257 	__atomic_store_n((p), (val), __ATOMIC_RELAXED)
258 
259 #endif /*COMPILER_H*/
260