1 /** 2 * \file common.h 3 * 4 * \brief Utility macros for internal use in the library 5 */ 6 /* 7 * Copyright The Mbed TLS Contributors 8 * SPDX-License-Identifier: Apache-2.0 OR GPL-2.0-or-later 9 */ 10 11 #ifndef MBEDTLS_LIBRARY_COMMON_H 12 #define MBEDTLS_LIBRARY_COMMON_H 13 14 #include "mbedtls/build_info.h" 15 #include "alignment.h" 16 17 #include <assert.h> 18 #include <stddef.h> 19 #include <stdint.h> 20 #include <stddef.h> 21 22 #if defined(__ARM_NEON) 23 /* 24 * Undefine and restore __section and __data from compiler.h to prevent 25 * collision with arm_neon.h 26 */ 27 #pragma push_macro("__section") 28 #pragma push_macro("__data") 29 #undef __section 30 #undef __data 31 #include <arm_neon.h> 32 #pragma pop_macro("__data") 33 #pragma pop_macro("__section") 34 #define MBEDTLS_HAVE_NEON_INTRINSICS 35 #elif defined(MBEDTLS_PLATFORM_IS_WINDOWS_ON_ARM64) 36 #include <arm64_neon.h> 37 #define MBEDTLS_HAVE_NEON_INTRINSICS 38 #endif 39 40 /** Helper to define a function as static except when building invasive tests. 41 * 42 * If a function is only used inside its own source file and should be 43 * declared `static` to allow the compiler to optimize for code size, 44 * but that function has unit tests, define it with 45 * ``` 46 * MBEDTLS_STATIC_TESTABLE int mbedtls_foo(...) { ... } 47 * ``` 48 * and declare it in a header in the `library/` directory with 49 * ``` 50 * #if defined(MBEDTLS_TEST_HOOKS) 51 * int mbedtls_foo(...); 52 * #endif 53 * ``` 54 */ 55 #if defined(MBEDTLS_TEST_HOOKS) 56 #define MBEDTLS_STATIC_TESTABLE 57 #else 58 #define MBEDTLS_STATIC_TESTABLE static 59 #endif 60 61 #if defined(MBEDTLS_TEST_HOOKS) 62 extern void (*mbedtls_test_hook_test_fail)(const char *test, int line, const char *file); 63 #define MBEDTLS_TEST_HOOK_TEST_ASSERT(TEST) \ 64 do { \ 65 if ((!(TEST)) && ((*mbedtls_test_hook_test_fail) != NULL)) \ 66 { \ 67 (*mbedtls_test_hook_test_fail)( #TEST, __LINE__, __FILE__); \ 68 } \ 69 } while (0) 70 #else 71 #define MBEDTLS_TEST_HOOK_TEST_ASSERT(TEST) 72 #endif /* defined(MBEDTLS_TEST_HOOKS) */ 73 74 /** \def ARRAY_LENGTH 75 * Return the number of elements of a static or stack array. 76 * 77 * \param array A value of array (not pointer) type. 78 * 79 * \return The number of elements of the array. 80 */ 81 /* A correct implementation of ARRAY_LENGTH, but which silently gives 82 * a nonsensical result if called with a pointer rather than an array. */ 83 #define ARRAY_LENGTH_UNSAFE(array) \ 84 (sizeof(array) / sizeof(*(array))) 85 86 #if defined(__GNUC__) 87 /* Test if arg and &(arg)[0] have the same type. This is true if arg is 88 * an array but not if it's a pointer. */ 89 #define IS_ARRAY_NOT_POINTER(arg) \ 90 (!__builtin_types_compatible_p(__typeof__(arg), \ 91 __typeof__(&(arg)[0]))) 92 /* A compile-time constant with the value 0. If `const_expr` is not a 93 * compile-time constant with a nonzero value, cause a compile-time error. */ 94 #define STATIC_ASSERT_EXPR(const_expr) \ 95 (0 && sizeof(struct { unsigned int STATIC_ASSERT : 1 - 2 * !(const_expr); })) 96 97 /* Return the scalar value `value` (possibly promoted). This is a compile-time 98 * constant if `value` is. `condition` must be a compile-time constant. 99 * If `condition` is false, arrange to cause a compile-time error. */ 100 #define STATIC_ASSERT_THEN_RETURN(condition, value) \ 101 (STATIC_ASSERT_EXPR(condition) ? 0 : (value)) 102 103 #define ARRAY_LENGTH(array) \ 104 (STATIC_ASSERT_THEN_RETURN(IS_ARRAY_NOT_POINTER(array), \ 105 ARRAY_LENGTH_UNSAFE(array))) 106 107 #else 108 /* If we aren't sure the compiler supports our non-standard tricks, 109 * fall back to the unsafe implementation. */ 110 #define ARRAY_LENGTH(array) ARRAY_LENGTH_UNSAFE(array) 111 #endif 112 /** Allow library to access its structs' private members. 113 * 114 * Although structs defined in header files are publicly available, 115 * their members are private and should not be accessed by the user. 116 */ 117 #define MBEDTLS_ALLOW_PRIVATE_ACCESS 118 119 /** 120 * \brief Securely zeroize a buffer then free it. 121 * 122 * Similar to making consecutive calls to 123 * \c mbedtls_platform_zeroize() and \c mbedtls_free(), but has 124 * code size savings, and potential for optimisation in the future. 125 * 126 * Guaranteed to be a no-op if \p buf is \c NULL and \p len is 0. 127 * 128 * \param buf Buffer to be zeroized then freed. 129 * \param len Length of the buffer in bytes 130 */ 131 void mbedtls_zeroize_and_free(void *buf, size_t len); 132 133 /** Return an offset into a buffer. 134 * 135 * This is just the addition of an offset to a pointer, except that this 136 * function also accepts an offset of 0 into a buffer whose pointer is null. 137 * (`p + n` has undefined behavior when `p` is null, even when `n == 0`. 138 * A null pointer is a valid buffer pointer when the size is 0, for example 139 * as the result of `malloc(0)` on some platforms.) 140 * 141 * \param p Pointer to a buffer of at least n bytes. 142 * This may be \p NULL if \p n is zero. 143 * \param n An offset in bytes. 144 * \return Pointer to offset \p n in the buffer \p p. 145 * Note that this is only a valid pointer if the size of the 146 * buffer is at least \p n + 1. 147 */ 148 static inline unsigned char *mbedtls_buffer_offset( 149 unsigned char *p, size_t n) 150 { 151 return p == NULL ? NULL : p + n; 152 } 153 154 /** Return an offset into a read-only buffer. 155 * 156 * Similar to mbedtls_buffer_offset(), but for const pointers. 157 * 158 * \param p Pointer to a buffer of at least n bytes. 159 * This may be \p NULL if \p n is zero. 160 * \param n An offset in bytes. 161 * \return Pointer to offset \p n in the buffer \p p. 162 * Note that this is only a valid pointer if the size of the 163 * buffer is at least \p n + 1. 164 */ 165 static inline const unsigned char *mbedtls_buffer_offset_const( 166 const unsigned char *p, size_t n) 167 { 168 return p == NULL ? NULL : p + n; 169 } 170 171 /* Always inline mbedtls_xor() for similar reasons as mbedtls_xor_no_simd(). */ 172 #if defined(__IAR_SYSTEMS_ICC__) 173 #pragma inline = forced 174 #elif defined(__GNUC__) 175 __attribute__((always_inline)) 176 #endif 177 /** 178 * Perform a fast block XOR operation, such that 179 * r[i] = a[i] ^ b[i] where 0 <= i < n 180 * 181 * \param r Pointer to result (buffer of at least \p n bytes). \p r 182 * may be equal to either \p a or \p b, but behaviour when 183 * it overlaps in other ways is undefined. 184 * \param a Pointer to input (buffer of at least \p n bytes) 185 * \param b Pointer to input (buffer of at least \p n bytes) 186 * \param n Number of bytes to process. 187 * 188 * \note Depending on the situation, it may be faster to use either mbedtls_xor() or 189 * mbedtls_xor_no_simd() (these are functionally equivalent). 190 * If the result is used immediately after the xor operation in non-SIMD code (e.g, in 191 * AES-CBC), there may be additional latency to transfer the data from SIMD to scalar 192 * registers, and in this case, mbedtls_xor_no_simd() may be faster. In other cases where 193 * the result is not used immediately (e.g., in AES-CTR), mbedtls_xor() may be faster. 194 * For targets without SIMD support, they will behave the same. 195 */ 196 static inline void mbedtls_xor(unsigned char *r, 197 const unsigned char *a, 198 const unsigned char *b, 199 size_t n) 200 { 201 size_t i = 0; 202 #if defined(MBEDTLS_EFFICIENT_UNALIGNED_ACCESS) 203 #if defined(MBEDTLS_HAVE_NEON_INTRINSICS) && \ 204 (!(defined(MBEDTLS_COMPILER_IS_GCC) && MBEDTLS_GCC_VERSION < 70300)) 205 /* Old GCC versions generate a warning here, so disable the NEON path for these compilers */ 206 for (; (i + 16) <= n; i += 16) { 207 uint8x16_t v1 = vld1q_u8(a + i); 208 uint8x16_t v2 = vld1q_u8(b + i); 209 uint8x16_t x = veorq_u8(v1, v2); 210 vst1q_u8(r + i, x); 211 } 212 #if defined(__IAR_SYSTEMS_ICC__) 213 /* This if statement helps some compilers (e.g., IAR) optimise out the byte-by-byte tail case 214 * where n is a constant multiple of 16. 215 * For other compilers (e.g. recent gcc and clang) it makes no difference if n is a compile-time 216 * constant, and is a very small perf regression if n is not a compile-time constant. */ 217 if (n % 16 == 0) { 218 return; 219 } 220 #endif 221 #elif defined(MBEDTLS_ARCH_IS_X64) || defined(MBEDTLS_ARCH_IS_ARM64) 222 /* This codepath probably only makes sense on architectures with 64-bit registers */ 223 for (; (i + 8) <= n; i += 8) { 224 uint64_t x = mbedtls_get_unaligned_uint64(a + i) ^ mbedtls_get_unaligned_uint64(b + i); 225 mbedtls_put_unaligned_uint64(r + i, x); 226 } 227 #if defined(__IAR_SYSTEMS_ICC__) 228 if (n % 8 == 0) { 229 return; 230 } 231 #endif 232 #else 233 for (; (i + 4) <= n; i += 4) { 234 uint32_t x = mbedtls_get_unaligned_uint32(a + i) ^ mbedtls_get_unaligned_uint32(b + i); 235 mbedtls_put_unaligned_uint32(r + i, x); 236 } 237 #if defined(__IAR_SYSTEMS_ICC__) 238 if (n % 4 == 0) { 239 return; 240 } 241 #endif 242 #endif 243 #endif 244 for (; i < n; i++) { 245 r[i] = a[i] ^ b[i]; 246 } 247 } 248 249 /* Always inline mbedtls_xor_no_simd() as we see significant perf regressions when it does not get 250 * inlined (e.g., observed about 3x perf difference in gcm_mult_largetable with gcc 7 - 12) */ 251 #if defined(__IAR_SYSTEMS_ICC__) 252 #pragma inline = forced 253 #elif defined(__GNUC__) 254 __attribute__((always_inline)) 255 #endif 256 /** 257 * Perform a fast block XOR operation, such that 258 * r[i] = a[i] ^ b[i] where 0 <= i < n 259 * 260 * In some situations, this can perform better than mbedtls_xor() (e.g., it's about 5% 261 * better in AES-CBC). 262 * 263 * \param r Pointer to result (buffer of at least \p n bytes). \p r 264 * may be equal to either \p a or \p b, but behaviour when 265 * it overlaps in other ways is undefined. 266 * \param a Pointer to input (buffer of at least \p n bytes) 267 * \param b Pointer to input (buffer of at least \p n bytes) 268 * \param n Number of bytes to process. 269 * 270 * \note Depending on the situation, it may be faster to use either mbedtls_xor() or 271 * mbedtls_xor_no_simd() (these are functionally equivalent). 272 * If the result is used immediately after the xor operation in non-SIMD code (e.g, in 273 * AES-CBC), there may be additional latency to transfer the data from SIMD to scalar 274 * registers, and in this case, mbedtls_xor_no_simd() may be faster. In other cases where 275 * the result is not used immediately (e.g., in AES-CTR), mbedtls_xor() may be faster. 276 * For targets without SIMD support, they will behave the same. 277 */ 278 static inline void mbedtls_xor_no_simd(unsigned char *r, 279 const unsigned char *a, 280 const unsigned char *b, 281 size_t n) 282 { 283 size_t i = 0; 284 #if defined(MBEDTLS_EFFICIENT_UNALIGNED_ACCESS) 285 #if defined(MBEDTLS_ARCH_IS_X64) || defined(MBEDTLS_ARCH_IS_ARM64) 286 /* This codepath probably only makes sense on architectures with 64-bit registers */ 287 for (; (i + 8) <= n; i += 8) { 288 uint64_t x = mbedtls_get_unaligned_uint64(a + i) ^ mbedtls_get_unaligned_uint64(b + i); 289 mbedtls_put_unaligned_uint64(r + i, x); 290 } 291 #if defined(__IAR_SYSTEMS_ICC__) 292 /* This if statement helps some compilers (e.g., IAR) optimise out the byte-by-byte tail case 293 * where n is a constant multiple of 8. 294 * For other compilers (e.g. recent gcc and clang) it makes no difference if n is a compile-time 295 * constant, and is a very small perf regression if n is not a compile-time constant. */ 296 if (n % 8 == 0) { 297 return; 298 } 299 #endif 300 #else 301 for (; (i + 4) <= n; i += 4) { 302 uint32_t x = mbedtls_get_unaligned_uint32(a + i) ^ mbedtls_get_unaligned_uint32(b + i); 303 mbedtls_put_unaligned_uint32(r + i, x); 304 } 305 #if defined(__IAR_SYSTEMS_ICC__) 306 if (n % 4 == 0) { 307 return; 308 } 309 #endif 310 #endif 311 #endif 312 for (; i < n; i++) { 313 r[i] = a[i] ^ b[i]; 314 } 315 } 316 317 /* Fix MSVC C99 compatible issue 318 * MSVC support __func__ from visual studio 2015( 1900 ) 319 * Use MSVC predefine macro to avoid name check fail. 320 */ 321 #if (defined(_MSC_VER) && (_MSC_VER <= 1900)) 322 #define /*no-check-names*/ __func__ __FUNCTION__ 323 #endif 324 325 /* Define `asm` for compilers which don't define it. */ 326 /* *INDENT-OFF* */ 327 #ifndef asm 328 #if defined(__IAR_SYSTEMS_ICC__) 329 #define asm __asm 330 #else 331 #define asm __asm__ 332 #endif 333 #endif 334 /* *INDENT-ON* */ 335 336 /* 337 * Define the constraint used for read-only pointer operands to aarch64 asm. 338 * 339 * This is normally the usual "r", but for aarch64_32 (aka ILP32, 340 * as found in watchos), "p" is required to avoid warnings from clang. 341 * 342 * Note that clang does not recognise '+p' or '=p', and armclang 343 * does not recognise 'p' at all. Therefore, to update a pointer from 344 * aarch64 assembly, it is necessary to use something like: 345 * 346 * uintptr_t uptr = (uintptr_t) ptr; 347 * asm( "ldr x4, [%x0], #8" ... : "+r" (uptr) : : ) 348 * ptr = (void*) uptr; 349 * 350 * Note that the "x" in "%x0" is neccessary; writing "%0" will cause warnings. 351 */ 352 #if defined(__aarch64__) && defined(MBEDTLS_HAVE_ASM) 353 #if UINTPTR_MAX == 0xfffffffful 354 /* ILP32: Specify the pointer operand slightly differently, as per #7787. */ 355 #define MBEDTLS_ASM_AARCH64_PTR_CONSTRAINT "p" 356 #elif UINTPTR_MAX == 0xfffffffffffffffful 357 /* Normal case (64-bit pointers): use "r" as the constraint for pointer operands to asm */ 358 #define MBEDTLS_ASM_AARCH64_PTR_CONSTRAINT "r" 359 #else 360 #error "Unrecognised pointer size for aarch64" 361 #endif 362 #endif 363 364 /* Always provide a static assert macro, so it can be used unconditionally. 365 * It will expand to nothing on some systems. 366 * Can be used outside functions (but don't add a trailing ';' in that case: 367 * the semicolon is included here to avoid triggering -Wextra-semi when 368 * MBEDTLS_STATIC_ASSERT() expands to nothing). 369 * Can't use the C11-style `defined(static_assert)` on FreeBSD, since it 370 * defines static_assert even with -std=c99, but then complains about it. 371 */ 372 #if defined(static_assert) && !defined(__FreeBSD__) 373 #define MBEDTLS_STATIC_ASSERT(expr, msg) static_assert(expr, msg); 374 #else 375 #define MBEDTLS_STATIC_ASSERT(expr, msg) 376 #endif 377 378 #if defined(__has_builtin) 379 #define MBEDTLS_HAS_BUILTIN(x) __has_builtin(x) 380 #else 381 #define MBEDTLS_HAS_BUILTIN(x) 0 382 #endif 383 384 /* Define compiler branch hints */ 385 #if MBEDTLS_HAS_BUILTIN(__builtin_expect) 386 #define MBEDTLS_LIKELY(x) __builtin_expect(!!(x), 1) 387 #define MBEDTLS_UNLIKELY(x) __builtin_expect(!!(x), 0) 388 #else 389 #define MBEDTLS_LIKELY(x) x 390 #define MBEDTLS_UNLIKELY(x) x 391 #endif 392 393 /* MBEDTLS_ASSUME may be used to provide additional information to the compiler 394 * which can result in smaller code-size. */ 395 #if MBEDTLS_HAS_BUILTIN(__builtin_assume) 396 /* clang provides __builtin_assume */ 397 #define MBEDTLS_ASSUME(x) __builtin_assume(x) 398 #elif MBEDTLS_HAS_BUILTIN(__builtin_unreachable) 399 /* gcc and IAR can use __builtin_unreachable */ 400 #define MBEDTLS_ASSUME(x) do { if (!(x)) __builtin_unreachable(); } while (0) 401 #elif defined(_MSC_VER) 402 /* Supported by MSVC since VS 2005 */ 403 #define MBEDTLS_ASSUME(x) __assume(x) 404 #else 405 #define MBEDTLS_ASSUME(x) do { } while (0) 406 #endif 407 408 /* For gcc -Os, override with -O2 for a given function. 409 * 410 * This will not affect behaviour for other optimisation settings, e.g. -O0. 411 */ 412 #if defined(MBEDTLS_COMPILER_IS_GCC) && defined(__OPTIMIZE_SIZE__) 413 #define MBEDTLS_OPTIMIZE_FOR_PERFORMANCE __attribute__((optimize("-O2"))) 414 #else 415 #define MBEDTLS_OPTIMIZE_FOR_PERFORMANCE 416 #endif 417 418 /* Suppress compiler warnings for unused functions and variables. */ 419 #if !defined(MBEDTLS_MAYBE_UNUSED) && defined(__has_attribute) 420 # if __has_attribute(unused) 421 # define MBEDTLS_MAYBE_UNUSED __attribute__((unused)) 422 # endif 423 #endif 424 #if !defined(MBEDTLS_MAYBE_UNUSED) && defined(__GNUC__) 425 # define MBEDTLS_MAYBE_UNUSED __attribute__((unused)) 426 #endif 427 #if !defined(MBEDTLS_MAYBE_UNUSED) && defined(__IAR_SYSTEMS_ICC__) && defined(__VER__) 428 /* IAR does support __attribute__((unused)), but only if the -e flag (extended language support) 429 * is given; the pragma always works. 430 * Unfortunately the pragma affects the rest of the file where it is used, but this is harmless. 431 * Check for version 5.2 or later - this pragma may be supported by earlier versions, but I wasn't 432 * able to find documentation). 433 */ 434 # if (__VER__ >= 5020000) 435 # define MBEDTLS_MAYBE_UNUSED _Pragma("diag_suppress=Pe177") 436 # endif 437 #endif 438 #if !defined(MBEDTLS_MAYBE_UNUSED) && defined(_MSC_VER) 439 # define MBEDTLS_MAYBE_UNUSED __pragma(warning(suppress:4189)) 440 #endif 441 #if !defined(MBEDTLS_MAYBE_UNUSED) 442 # define MBEDTLS_MAYBE_UNUSED 443 #endif 444 445 #endif /* MBEDTLS_LIBRARY_COMMON_H */ 446