aes_neon.c revision 1.1
1 /* $NetBSD: aes_neon.c,v 1.1 2020/06/29 23:56:31 riastradh Exp $ */ 2 3 /*- 4 * Copyright (c) 2020 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 17 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 18 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 19 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 20 * BE 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 /* 30 * Permutation-based AES using NEON, derived from Mike Hamburg's VPAES 31 * software, at <https://crypto.stanford.edu/vpaes/>, described in 32 * 33 * Mike Hamburg, `Accelerating AES with Vector Permute 34 * Instructions', in Christophe Clavier and Kris Gaj (eds.), 35 * Cryptographic Hardware and Embedded Systems -- CHES 2009, 36 * Springer LNCS 5747, pp. 18-32. 37 * 38 * https://link.springer.com/chapter/10.1007/978-3-642-04138-9_2 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(1, "$NetBSD: aes_neon.c,v 1.1 2020/06/29 23:56:31 riastradh Exp $"); 43 44 #include <sys/types.h> 45 46 #include <sys/systm.h> 47 48 #include "aes_neon_impl.h" 49 50 static const uint8x16_t 51 mc_forward[4] = { 52 {0x01,0x02,0x03,0x00,0x05,0x06,0x07,0x04, 53 0x09,0x0A,0x0B,0x08,0x0D,0x0E,0x0F,0x0C}, 54 {0x05,0x06,0x07,0x04,0x09,0x0A,0x0B,0x08, 55 0x0D,0x0E,0x0F,0x0C,0x01,0x02,0x03,0x00}, 56 {0x09,0x0A,0x0B,0x08,0x0D,0x0E,0x0F,0x0C, 57 0x01,0x02,0x03,0x00,0x05,0x06,0x07,0x04}, 58 {0x0D,0x0E,0x0F,0x0C,0x01,0x02,0x03,0x00, 59 0x05,0x06,0x07,0x04,0x09,0x0A,0x0B,0x08}, 60 }, 61 mc_backward[4] = { 62 {0x03,0x00,0x01,0x02,0x07,0x04,0x05,0x06, 63 0x0B,0x08,0x09,0x0A,0x0F,0x0C,0x0D,0x0E}, 64 {0x0F,0x0C,0x0D,0x0E,0x03,0x00,0x01,0x02, 65 0x07,0x04,0x05,0x06,0x0B,0x08,0x09,0x0A}, 66 {0x0B,0x08,0x09,0x0A,0x0F,0x0C,0x0D,0x0E, 67 0x03,0x00,0x01,0x02,0x07,0x04,0x05,0x06}, 68 {0x07,0x04,0x05,0x06,0x0B,0x08,0x09,0x0A, 69 0x0F,0x0C,0x0D,0x0E,0x03,0x00,0x01,0x02}, 70 }, 71 ipt[2] = { 72 {0x00,0x70,0x2A,0x5A,0x98,0xE8,0xB2,0xC2, 73 0x08,0x78,0x22,0x52,0x90,0xE0,0xBA,0xCA}, 74 {0x00,0x4D,0x7C,0x31,0x7D,0x30,0x01,0x4C, 75 0x81,0xCC,0xFD,0xB0,0xFC,0xB1,0x80,0xCD}, 76 }, 77 opt[2] = { 78 {0x00,0x60,0xB6,0xD6,0x29,0x49,0x9F,0xFF, 79 0x08,0x68,0xBE,0xDE,0x21,0x41,0x97,0xF7}, 80 {0x00,0xEC,0xBC,0x50,0x51,0xBD,0xED,0x01, 81 0xE0,0x0C,0x5C,0xB0,0xB1,0x5D,0x0D,0xE1}, 82 }, 83 dipt[2] = { 84 {0x00,0x5F,0x54,0x0B,0x04,0x5B,0x50,0x0F, 85 0x1A,0x45,0x4E,0x11,0x1E,0x41,0x4A,0x15}, 86 {0x00,0x65,0x05,0x60,0xE6,0x83,0xE3,0x86, 87 0x94,0xF1,0x91,0xF4,0x72,0x17,0x77,0x12}, 88 }, 89 sb1[2] = { 90 {0x00,0x3E,0x50,0xCB,0x8F,0xE1,0x9B,0xB1, 91 0x44,0xF5,0x2A,0x14,0x6E,0x7A,0xDF,0xA5}, 92 {0x00,0x23,0xE2,0xFA,0x15,0xD4,0x18,0x36, 93 0xEF,0xD9,0x2E,0x0D,0xC1,0xCC,0xF7,0x3B}, 94 }, 95 sb2[2] = { 96 {0x00,0x24,0x71,0x0B,0xC6,0x93,0x7A,0xE2, 97 0xCD,0x2F,0x98,0xBC,0x55,0xE9,0xB7,0x5E}, 98 {0x00,0x29,0xE1,0x0A,0x40,0x88,0xEB,0x69, 99 0x4A,0x23,0x82,0xAB,0xC8,0x63,0xA1,0xC2}, 100 }, 101 sbo[2] = { 102 {0x00,0xC7,0xBD,0x6F,0x17,0x6D,0xD2,0xD0, 103 0x78,0xA8,0x02,0xC5,0x7A,0xBF,0xAA,0x15}, 104 {0x00,0x6A,0xBB,0x5F,0xA5,0x74,0xE4,0xCF, 105 0xFA,0x35,0x2B,0x41,0xD1,0x90,0x1E,0x8E}, 106 }, 107 dsb9[2] = { 108 {0x00,0xD6,0x86,0x9A,0x53,0x03,0x1C,0x85, 109 0xC9,0x4C,0x99,0x4F,0x50,0x1F,0xD5,0xCA}, 110 {0x00,0x49,0xD7,0xEC,0x89,0x17,0x3B,0xC0, 111 0x65,0xA5,0xFB,0xB2,0x9E,0x2C,0x5E,0x72}, 112 }, 113 dsbd[2] = { 114 {0x00,0xA2,0xB1,0xE6,0xDF,0xCC,0x57,0x7D, 115 0x39,0x44,0x2A,0x88,0x13,0x9B,0x6E,0xF5}, 116 {0x00,0xCB,0xC6,0x24,0xF7,0xFA,0xE2,0x3C, 117 0xD3,0xEF,0xDE,0x15,0x0D,0x18,0x31,0x29}, 118 }, 119 dsbb[2] = { 120 {0x00,0x42,0xB4,0x96,0x92,0x64,0x22,0xD0, 121 0x04,0xD4,0xF2,0xB0,0xF6,0x46,0x26,0x60}, 122 {0x00,0x67,0x59,0xCD,0xA6,0x98,0x94,0xC1, 123 0x6B,0xAA,0x55,0x32,0x3E,0x0C,0xFF,0xF3}, 124 }, 125 dsbe[2] = { 126 {0x00,0xD0,0xD4,0x26,0x96,0x92,0xF2,0x46, 127 0xB0,0xF6,0xB4,0x64,0x04,0x60,0x42,0x22}, 128 {0x00,0xC1,0xAA,0xFF,0xCD,0xA6,0x55,0x0C, 129 0x32,0x3E,0x59,0x98,0x6B,0xF3,0x67,0x94}, 130 }, 131 dsbo[2] = { 132 {0x00,0x40,0xF9,0x7E,0x53,0xEA,0x87,0x13, 133 0x2D,0x3E,0x94,0xD4,0xB9,0x6D,0xAA,0xC7}, 134 {0x00,0x1D,0x44,0x93,0x0F,0x56,0xD7,0x12, 135 0x9C,0x8E,0xC5,0xD8,0x59,0x81,0x4B,0xCA}, 136 }, 137 dks1[2] = { 138 {0x00,0xA7,0xD9,0x7E,0xC8,0x6F,0x11,0xB6, 139 0xFC,0x5B,0x25,0x82,0x34,0x93,0xED,0x4A}, 140 {0x00,0x33,0x14,0x27,0x62,0x51,0x76,0x45, 141 0xCE,0xFD,0xDA,0xE9,0xAC,0x9F,0xB8,0x8B}, 142 }, 143 dks2[2] = { 144 {0x00,0x64,0xA8,0xCC,0xEB,0x8F,0x43,0x27, 145 0x61,0x05,0xC9,0xAD,0x8A,0xEE,0x22,0x46}, 146 {0x00,0xDD,0x92,0x4F,0xCE,0x13,0x5C,0x81, 147 0xF2,0x2F,0x60,0xBD,0x3C,0xE1,0xAE,0x73}, 148 }, 149 dks3[2] = { 150 {0x00,0xC7,0xC6,0x01,0x02,0xC5,0xC4,0x03, 151 0xFB,0x3C,0x3D,0xFA,0xF9,0x3E,0x3F,0xF8}, 152 {0x00,0xF7,0xCF,0x38,0xD6,0x21,0x19,0xEE, 153 0x4B,0xBC,0x84,0x73,0x9D,0x6A,0x52,0xA5}, 154 }, 155 dks4[2] = { 156 {0x00,0x20,0x73,0x53,0xB0,0x90,0xC3,0xE3, 157 0x43,0x63,0x30,0x10,0xF3,0xD3,0x80,0xA0}, 158 {0xE8,0x82,0x69,0x03,0x4B,0x21,0xCA,0xA0, 159 0x67,0x0D,0xE6,0x8C,0xC4,0xAE,0x45,0x2F}, 160 }, 161 deskew[2] = { 162 {0x00,0xE3,0xA4,0x47,0x40,0xA3,0xE4,0x07, 163 0x1A,0xF9,0xBE,0x5D,0x5A,0xB9,0xFE,0x1D}, 164 {0x00,0x69,0xEA,0x83,0xDC,0xB5,0x36,0x5F, 165 0x77,0x1E,0x9D,0xF4,0xAB,0xC2,0x41,0x28}, 166 }, 167 sr[4] = { 168 {0x00,0x01,0x02,0x03,0x04,0x05,0x06,0x07, 169 0x08,0x09,0x0A,0x0B,0x0C,0x0D,0x0E,0x0F}, 170 {0x00,0x05,0x0A,0x0F,0x04,0x09,0x0E,0x03, 171 0x08,0x0D,0x02,0x07,0x0C,0x01,0x06,0x0B}, 172 {0x00,0x09,0x02,0x0B,0x04,0x0D,0x06,0x0F, 173 0x08,0x01,0x0A,0x03,0x0C,0x05,0x0E,0x07}, 174 {0x00,0x0D,0x0A,0x07,0x04,0x01,0x0E,0x0B, 175 0x08,0x05,0x02,0x0F,0x0C,0x09,0x06,0x03}, 176 }, 177 rcon = {0xB6,0xEE,0x9D,0xAF,0xB9,0x91,0x83,0x1F, 178 0x81,0x7D,0x7C,0x4D,0x08,0x98,0x2A,0x70}, 179 s63 = {0x5B,0x5B,0x5B,0x5B,0x5B,0x5B,0x5B,0x5B, 180 0x5B,0x5B,0x5B,0x5B,0x5B,0x5B,0x5B,0x5B}, 181 of = {0x0F,0x0F,0x0F,0x0F,0x0F,0x0F,0x0F,0x0F, 182 0x0F,0x0F,0x0F,0x0F,0x0F,0x0F,0x0F,0x0F}, 183 inv = {0x80,0x01,0x08,0x0D,0x0F,0x06,0x05,0x0E, 184 0x02,0x0C,0x0B,0x0A,0x09,0x03,0x07,0x04}, 185 inva = {0x80,0x07,0x0B,0x0F,0x06,0x0A,0x04,0x01, 186 0x09,0x08,0x05,0x02,0x0C,0x0E,0x0D,0x03}; 187 188 static inline uint8x16_t 189 loadroundkey(const void *rkp) 190 { 191 return vld1q_u8(rkp); 192 } 193 194 static inline void 195 storeroundkey(void *rkp, uint8x16_t rk) 196 { 197 vst1q_u8(rkp, rk); 198 } 199 200 /* Given abcdefgh, set *lo = 0b0d0f0h and *hi = 0a0c0e0g. */ 201 static inline void 202 bytes2nybbles(uint8x16_t *restrict lo, uint8x16_t *restrict hi, uint8x16_t x) 203 { 204 205 *lo = of & x; 206 *hi = of & vshrq_n_u8(x, 4); 207 } 208 209 /* 210 * t is a pair of maps respectively from low and high nybbles to bytes. 211 * Apply t the nybbles, and add the results in GF(2). 212 */ 213 static uint8x16_t 214 aes_schedule_transform(uint8x16_t x, const uint8x16_t t[static 2]) 215 { 216 uint8x16_t lo, hi; 217 218 bytes2nybbles(&lo, &hi, x); 219 return vqtbl1q_u8(t[0], lo) ^ vqtbl1q_u8(t[1], hi); 220 } 221 222 static inline void 223 subbytes(uint8x16_t *io, uint8x16_t *jo, uint8x16_t x, uint8x16_t inv_, 224 uint8x16_t inva_) 225 { 226 uint8x16_t k, i, ak, j; 227 228 bytes2nybbles(&k, &i, x); 229 ak = vqtbl1q_u8(inva_, k); 230 j = i ^ k; 231 *io = j ^ vqtbl1q_u8(inv_, ak ^ vqtbl1q_u8(inv_, i)); 232 *jo = i ^ vqtbl1q_u8(inv_, ak ^ vqtbl1q_u8(inv_, j)); 233 } 234 235 static uint8x16_t 236 aes_schedule_low_round(uint8x16_t rk, uint8x16_t prk) 237 { 238 uint8x16_t io, jo; 239 240 /* smear prk */ 241 prk ^= vextq_u8(vdupq_n_u8(0), prk, 12); 242 prk ^= vextq_u8(vdupq_n_u8(0), prk, 8); 243 prk ^= s63; 244 245 /* subbytes */ 246 subbytes(&io, &jo, rk, inv, inva); 247 rk = vqtbl1q_u8(sb1[0], io) ^ vqtbl1q_u8(sb1[1], jo); 248 249 /* add in smeared stuff */ 250 return rk ^ prk; 251 } 252 253 static uint8x16_t 254 aes_schedule_round(uint8x16_t rk, uint8x16_t prk, uint8x16_t *rcon_rot) 255 { 256 uint32x4_t rk32; 257 258 /* extract rcon from rcon_rot */ 259 prk ^= vextq_u8(*rcon_rot, vdupq_n_u8(0), 15); 260 *rcon_rot = vextq_u8(*rcon_rot, *rcon_rot, 15); 261 262 /* rotate */ 263 rk32 = vreinterpretq_u32_u8(rk); 264 rk32 = vdupq_n_u32(vgetq_lane_u32(rk32, 3)); 265 rk = vreinterpretq_u8_u32(rk32); 266 rk = vextq_u8(rk, rk, 1); 267 268 return aes_schedule_low_round(rk, prk); 269 } 270 271 static uint8x16_t 272 aes_schedule_mangle_enc(uint8x16_t x, uint8x16_t sr_i) 273 { 274 uint8x16_t y = vdupq_n_u8(0); 275 276 x ^= s63; 277 278 x = vqtbl1q_u8(x, mc_forward[0]); 279 y ^= x; 280 x = vqtbl1q_u8(x, mc_forward[0]); 281 y ^= x; 282 x = vqtbl1q_u8(x, mc_forward[0]); 283 y ^= x; 284 285 return vqtbl1q_u8(y, sr_i); 286 } 287 288 static uint8x16_t 289 aes_schedule_mangle_last_enc(uint8x16_t x, uint8x16_t sr_i) 290 { 291 292 return aes_schedule_transform(vqtbl1q_u8(x, sr_i) ^ s63, opt); 293 } 294 295 static uint8x16_t 296 aes_schedule_mangle_dec(uint8x16_t x, uint8x16_t sr_i) 297 { 298 uint8x16_t y = vdupq_n_u8(0); 299 300 x = aes_schedule_transform(x, dks1); 301 y = vqtbl1q_u8(y ^ x, mc_forward[0]); 302 x = aes_schedule_transform(x, dks2); 303 y = vqtbl1q_u8(y ^ x, mc_forward[0]); 304 x = aes_schedule_transform(x, dks3); 305 y = vqtbl1q_u8(y ^ x, mc_forward[0]); 306 x = aes_schedule_transform(x, dks4); 307 y = vqtbl1q_u8(y ^ x, mc_forward[0]); 308 309 return vqtbl1q_u8(y, sr_i); 310 } 311 312 static uint8x16_t 313 aes_schedule_mangle_last_dec(uint8x16_t x) 314 { 315 316 return aes_schedule_transform(x ^ s63, deskew); 317 } 318 319 static uint8x16_t 320 aes_schedule_192_smear(uint8x16_t prkhi, uint8x16_t prk) 321 { 322 uint32x4_t prkhi32 = vreinterpretq_u32_u8(prkhi); 323 uint32x4_t prk32 = vreinterpretq_u32_u8(prk); 324 uint32x4_t rk32; 325 326 rk32 = prkhi32; 327 rk32 ^= vsetq_lane_u32(vgetq_lane_u32(prkhi32, 2), 328 vdupq_n_u32(vgetq_lane_u32(prkhi32, 0)), 329 3); 330 rk32 ^= vsetq_lane_u32(vgetq_lane_u32(prk32, 2), 331 vdupq_n_u32(vgetq_lane_u32(prk32, 3)), 332 0); 333 334 return vreinterpretq_u8_u32(rk32); 335 } 336 337 static uint8x16_t 338 aes_schedule_192_smearhi(uint8x16_t rk) 339 { 340 uint64x2_t rk64 = vreinterpretq_u64_u8(rk); 341 342 rk64 = vsetq_lane_u64(0, rk64, 0); 343 344 return vreinterpretq_u8_u64(rk64); 345 } 346 347 void 348 aes_neon_setenckey(struct aesenc *enc, const uint8_t *key, unsigned nrounds) 349 { 350 uint32_t *rk32 = enc->aese_aes.aes_rk; 351 uint8x16_t mrk; /* mangled round key */ 352 uint8x16_t rk; /* round key */ 353 uint8x16_t prk; /* previous round key */ 354 uint8x16_t rcon_rot = rcon; 355 uint64_t i = 3; 356 357 /* input transform */ 358 rk = aes_schedule_transform(vld1q_u8(key), ipt); 359 storeroundkey(rk32, rk); 360 rk32 += 4; 361 362 switch (nrounds) { 363 case 10: 364 for (;;) { 365 rk = aes_schedule_round(rk, rk, &rcon_rot); 366 if (--nrounds == 0) 367 break; 368 mrk = aes_schedule_mangle_enc(rk, sr[i-- % 4]); 369 storeroundkey(rk32, mrk); 370 rk32 += 4; 371 } 372 break; 373 case 12: { 374 uint8x16_t prkhi; /* high half of previous round key */ 375 376 prk = rk; 377 rk = aes_schedule_transform(vld1q_u8(key + 8), ipt); 378 prkhi = aes_schedule_192_smearhi(rk); 379 for (;;) { 380 prk = aes_schedule_round(rk, prk, &rcon_rot); 381 rk = vextq_u8(prkhi, prk, 8); 382 383 mrk = aes_schedule_mangle_enc(rk, sr[i-- % 4]); 384 storeroundkey(rk32, mrk); 385 rk32 += 4; 386 rk = aes_schedule_192_smear(prkhi, prk); 387 prkhi = aes_schedule_192_smearhi(rk); 388 389 mrk = aes_schedule_mangle_enc(rk, sr[i-- % 4]); 390 storeroundkey(rk32, mrk); 391 rk32 += 4; 392 rk = prk = aes_schedule_round(rk, prk, &rcon_rot); 393 if ((nrounds -= 3) == 0) 394 break; 395 396 mrk = aes_schedule_mangle_enc(rk, sr[i-- % 4]); 397 storeroundkey(rk32, mrk); 398 rk32 += 4; 399 rk = aes_schedule_192_smear(prkhi, prk); 400 prkhi = aes_schedule_192_smearhi(rk); 401 } 402 break; 403 } 404 case 14: { 405 uint8x16_t pprk; /* previous previous round key */ 406 407 prk = rk; 408 rk = aes_schedule_transform(vld1q_u8(key + 16), ipt); 409 for (;;) { 410 mrk = aes_schedule_mangle_enc(rk, sr[i-- % 4]); 411 storeroundkey(rk32, mrk); 412 rk32 += 4; 413 pprk = rk; 414 415 /* high round */ 416 rk = prk = aes_schedule_round(rk, prk, &rcon_rot); 417 if ((nrounds -= 2) == 0) 418 break; 419 mrk = aes_schedule_mangle_enc(rk, sr[i-- % 4]); 420 storeroundkey(rk32, mrk); 421 rk32 += 4; 422 423 /* low round */ 424 rk = vreinterpretq_u8_u32( 425 vdupq_n_u32( 426 vgetq_lane_u32(vreinterpretq_u32_u8(rk), 427 3))); 428 rk = aes_schedule_low_round(rk, pprk); 429 } 430 break; 431 } 432 default: 433 panic("invalid number of AES rounds: %u", nrounds); 434 } 435 storeroundkey(rk32, aes_schedule_mangle_last_enc(rk, sr[i-- % 4])); 436 } 437 438 void 439 aes_neon_setdeckey(struct aesdec *dec, const uint8_t *key, unsigned nrounds) 440 { 441 uint32_t *rk32 = dec->aesd_aes.aes_rk; 442 uint8x16_t mrk; /* mangled round key */ 443 uint8x16_t ork; /* original round key */ 444 uint8x16_t rk; /* round key */ 445 uint8x16_t prk; /* previous round key */ 446 uint8x16_t rcon_rot = rcon; 447 unsigned i = nrounds == 12 ? 0 : 2; 448 449 ork = vld1q_u8(key); 450 451 /* input transform */ 452 rk = aes_schedule_transform(ork, ipt); 453 454 /* go from end */ 455 rk32 += 4*nrounds; 456 storeroundkey(rk32, vqtbl1q_u8(ork, sr[i])); 457 rk32 -= 4; 458 i ^= 3; 459 460 switch (nrounds) { 461 case 10: 462 for (;;) { 463 rk = aes_schedule_round(rk, rk, &rcon_rot); 464 if (--nrounds == 0) 465 break; 466 mrk = aes_schedule_mangle_dec(rk, sr[i-- % 4]); 467 storeroundkey(rk32, mrk); 468 rk32 -= 4; 469 } 470 break; 471 case 12: { 472 uint8x16_t prkhi; /* high half of previous round key */ 473 474 prk = rk; 475 rk = aes_schedule_transform(vld1q_u8(key + 8), ipt); 476 prkhi = aes_schedule_192_smearhi(rk); 477 for (;;) { 478 prk = aes_schedule_round(rk, prk, &rcon_rot); 479 rk = vextq_u8(prkhi, prk, 8); 480 481 mrk = aes_schedule_mangle_dec(rk, sr[i-- % 4]); 482 storeroundkey(rk32, mrk); 483 rk32 -= 4; 484 rk = aes_schedule_192_smear(prkhi, prk); 485 prkhi = aes_schedule_192_smearhi(rk); 486 487 mrk = aes_schedule_mangle_dec(rk, sr[i-- % 4]); 488 storeroundkey(rk32, mrk); 489 rk32 -= 4; 490 rk = prk = aes_schedule_round(rk, prk, &rcon_rot); 491 if ((nrounds -= 3) == 0) 492 break; 493 494 mrk = aes_schedule_mangle_dec(rk, sr[i-- % 4]); 495 storeroundkey(rk32, mrk); 496 rk32 -= 4; 497 rk = aes_schedule_192_smear(prkhi, prk); 498 prkhi = aes_schedule_192_smearhi(rk); 499 } 500 break; 501 } 502 case 14: { 503 uint8x16_t pprk; /* previous previous round key */ 504 505 prk = rk; 506 rk = aes_schedule_transform(vld1q_u8(key + 16), ipt); 507 for (;;) { 508 mrk = aes_schedule_mangle_dec(rk, sr[i-- % 4]); 509 storeroundkey(rk32, mrk); 510 rk32 -= 4; 511 pprk = rk; 512 513 /* high round */ 514 rk = prk = aes_schedule_round(rk, prk, &rcon_rot); 515 if ((nrounds -= 2) == 0) 516 break; 517 mrk = aes_schedule_mangle_dec(rk, sr[i-- % 4]); 518 storeroundkey(rk32, mrk); 519 rk32 -= 4; 520 521 /* low round */ 522 rk = vreinterpretq_u8_u32( 523 vdupq_n_u32( 524 vgetq_lane_u32(vreinterpretq_u32_u8(rk), 525 3))); 526 rk = aes_schedule_low_round(rk, pprk); 527 } 528 break; 529 } 530 default: 531 panic("invalid number of AES rounds: %u", nrounds); 532 } 533 storeroundkey(rk32, aes_schedule_mangle_last_dec(rk)); 534 } 535 536 uint8x16_t 537 aes_neon_enc1(const struct aesenc *enc, uint8x16_t x, unsigned nrounds) 538 { 539 const uint32_t *rk32 = enc->aese_aes.aes_rk; 540 uint8x16_t inv_ = *(const volatile uint8x16_t *)&inv; 541 uint8x16_t inva_ = *(const volatile uint8x16_t *)&inva; 542 uint8x16_t sb1_0 = ((const volatile uint8x16_t *)sb1)[0]; 543 uint8x16_t sb1_1 = ((const volatile uint8x16_t *)sb1)[1]; 544 uint8x16_t sb2_0 = ((const volatile uint8x16_t *)sb2)[0]; 545 uint8x16_t sb2_1 = ((const volatile uint8x16_t *)sb2)[1]; 546 uint8x16_t io, jo; 547 unsigned rmod4 = 0; 548 549 x = aes_schedule_transform(x, ipt); 550 x ^= loadroundkey(rk32); 551 for (;;) { 552 uint8x16_t A, A2, A2_B, A2_B_D; 553 554 subbytes(&io, &jo, x, inv_, inva_); 555 556 rk32 += 4; 557 rmod4 = (rmod4 + 1) % 4; 558 if (--nrounds == 0) 559 break; 560 561 A = vqtbl1q_u8(sb1_0, io) ^ vqtbl1q_u8(sb1_1, jo); 562 A ^= loadroundkey(rk32); 563 A2 = vqtbl1q_u8(sb2_0, io) ^ vqtbl1q_u8(sb2_1, jo); 564 A2_B = A2 ^ vqtbl1q_u8(A, mc_forward[rmod4]); 565 A2_B_D = A2_B ^ vqtbl1q_u8(A, mc_backward[rmod4]); 566 x = A2_B_D ^ vqtbl1q_u8(A2_B, mc_forward[rmod4]); 567 } 568 x = vqtbl1q_u8(sbo[0], io) ^ vqtbl1q_u8(sbo[1], jo); 569 x ^= loadroundkey(rk32); 570 return vqtbl1q_u8(x, sr[rmod4]); 571 } 572 573 uint8x16_t 574 aes_neon_dec1(const struct aesdec *dec, uint8x16_t x, unsigned nrounds) 575 { 576 const uint32_t *rk32 = dec->aesd_aes.aes_rk; 577 unsigned i = 3 & ~(nrounds - 1); 578 uint8x16_t inv_ = *(const volatile uint8x16_t *)&inv; 579 uint8x16_t inva_ = *(const volatile uint8x16_t *)&inva; 580 uint8x16_t io, jo, mc; 581 582 x = aes_schedule_transform(x, dipt); 583 x ^= loadroundkey(rk32); 584 rk32 += 4; 585 586 mc = mc_forward[3]; 587 for (;;) { 588 subbytes(&io, &jo, x, inv_, inva_); 589 if (--nrounds == 0) 590 break; 591 592 x = vqtbl1q_u8(dsb9[0], io) ^ vqtbl1q_u8(dsb9[1], jo); 593 x ^= loadroundkey(rk32); 594 rk32 += 4; /* next round key */ 595 596 x = vqtbl1q_u8(x, mc); 597 x ^= vqtbl1q_u8(dsbd[0], io) ^ vqtbl1q_u8(dsbd[1], jo); 598 599 x = vqtbl1q_u8(x, mc); 600 x ^= vqtbl1q_u8(dsbb[0], io) ^ vqtbl1q_u8(dsbb[1], jo); 601 602 x = vqtbl1q_u8(x, mc); 603 x ^= vqtbl1q_u8(dsbe[0], io) ^ vqtbl1q_u8(dsbe[1], jo); 604 605 mc = vextq_u8(mc, mc, 12); 606 } 607 x = vqtbl1q_u8(dsbo[0], io) ^ vqtbl1q_u8(dsbo[1], jo); 608 x ^= loadroundkey(rk32); 609 return vqtbl1q_u8(x, sr[i]); 610 } 611
Indexes created Tue Sep 30 17:09:57 GMT 2025