| | 1 | /*\r |
| | 2 | ---------------------------------------------------------------------------\r |
| | 3 | Copyright (c) 2003, Dr Brian Gladman, Worcester, UK. All rights reserved.\r |
| | 4 | \r |
| | 5 | LICENSE TERMS\r |
| | 6 | \r |
| | 7 | The free distribution and use of this software in both source and binary\r |
| | 8 | form is allowed (with or without changes) provided that:\r |
| | 9 | \r |
| | 10 | 1. distributions of this source code include the above copyright\r |
| | 11 | notice, this list of conditions and the following disclaimer;\r |
| | 12 | \r |
| | 13 | 2. distributions in binary form include the above copyright\r |
| | 14 | notice, this list of conditions and the following disclaimer\r |
| | 15 | in the documentation and/or other associated materials;\r |
| | 16 | \r |
| | 17 | 3. the copyright holder's name is not used to endorse products\r |
| | 18 | built using this software without specific written permission.\r |
| | 19 | \r |
| | 20 | ALTERNATIVELY, provided that this notice is retained in full, this product\r |
| | 21 | may be distributed under the terms of the GNU General Public License (GPL),\r |
| | 22 | in which case the provisions of the GPL apply INSTEAD OF those given above.\r |
| | 23 | \r |
| | 24 | DISCLAIMER\r |
| | 25 | \r |
| | 26 | This software is provided 'as is' with no explicit or implied warranties\r |
| | 27 | in respect of its properties, including, but not limited to, correctness\r |
| | 28 | and/or fitness for purpose.\r |
| | 29 | ---------------------------------------------------------------------------\r |
| | 30 | Issue 31/01/2006\r |
| | 31 | \r |
| | 32 | These subroutines implement multiple block AES modes for ECB, CBC, CFB,\r |
| | 33 | OFB and CTR encryption, The code provides support for the VIA Advanced \r |
| | 34 | Cryptography Engine (ACE).\r |
| | 35 | \r |
| | 36 | NOTE: In the following subroutines, the AES contexts (ctx) must be\r |
| | 37 | 16 byte aligned if VIA ACE is being used\r |
| | 38 | */\r |
| | 39 | \r |
| | 40 | \r |
| | 41 | /* ---------------------------------------------------------------------------------------------------------------- \r |
| | 42 | \r |
| | 43 | aes_encrypt_cbc function (see aes_modes.c or aes_modes_asm.s) :\r |
| | 44 | \r |
| | 45 | For simplicity, I am assuming all variables are in 128-bit data type.\r |
| | 46 | \r |
| | 47 | aes_rval aes_encrypt_cbc(const __m128 *ibuf, __m128 *iv, int num_blk, __m128 *obuf, const aes_encrypt_ctx *ctx)\r |
| | 48 | {\r |
| | 49 | while(num_blk--) {\r |
| | 50 | *iv ^= *ibuf++;\r |
| | 51 | aes_encrypt(iv, iv, ctx);\r |
| | 52 | *obuf++ = *iv;\r |
| | 53 | }\r |
| | 54 | return 0;\r |
| | 55 | }\r |
| | 56 | \r |
| | 57 | The following is an implementation of this function using Intel AESNI.\r |
| | 58 | This function _aes_encrypt_cbc_hw SHOULD NOT be called directly. \r |
| | 59 | Developer should still call _aes_encrypt_cbc (in aes_modes_asm.s) which will poll cpu_capabilities and branch\r |
| | 60 | to this aesni-based function should it detecs that aesni is available.\r |
| | 61 | Blindly call this function SURELY will cause a CRASH on systems with no aesni support. \r |
| | 62 | \r |
| | 63 | Note that each block starts with *iv, which is the output of the previous block. Therefore, the cbc blocks\r |
| | 64 | are serially chained. This prevents us from arranging several blocks for encryption in parallel.\r |
| | 65 | \r |
| | 66 | ----------------------------------------------------------------------------------------------------------------*/\r |
| | 67 | \r |
| | 68 | .text\r |
| | 69 | .align 4,0x90\r |
| | 70 | .globl _aes_encrypt_cbc_hw\r |
| | 71 | _aes_encrypt_cbc_hw:\r |
| | 72 | \r |
| | 73 | // push/save registers for local use\r |
| | 74 | #if defined __i386__\r |
| | 75 | \r |
| | 76 | push %ebp\r |
| | 77 | movl %esp, %ebp\r |
| | 78 | push %ebx\r |
| | 79 | push %edi\r |
| | 80 | \r |
| | 81 | #define sp %esp\r |
| | 82 | \r |
| | 83 | #else // __x86_64__\r |
| | 84 | \r |
| | 85 | push %rbp\r |
| | 86 | mov %rsp, %rbp\r |
| | 87 | push %rbx\r |
| | 88 | push %r13\r |
| | 89 | push %r14\r |
| | 90 | push %r15\r |
| | 91 | \r |
| | 92 | #define sp %rsp\r |
| | 93 | \r |
| | 94 | #endif\r |
| | 95 | \r |
| | 96 | // if this is kernel code, need to save used xmm registers\r |
| | 97 | #ifdef KERNEL\r |
| | 98 | \r |
| | 99 | #if defined __i386__\r |
| | 100 | sub $(8*16), %esp // for possible xmm0-xmm7 save/restore\r |
| | 101 | #else\r |
| | 102 | sub $(16*16), %rsp // xmm0-xmm15 save/restore \r |
| | 103 | #endif\r |
| | 104 | \r |
| | 105 | movaps %xmm0, (sp)\r |
| | 106 | movaps %xmm1, 16(sp)\r |
| | 107 | movaps %xmm2, 32(sp)\r |
| | 108 | movaps %xmm3, 48(sp)\r |
| | 109 | movaps %xmm4, 64(sp)\r |
| | 110 | movaps %xmm5, 80(sp)\r |
| | 111 | movaps %xmm6, 96(sp)\r |
| | 112 | movaps %xmm7, 112(sp)\r |
| | 113 | #if defined __x86_64__\r |
| | 114 | movaps %xmm8, 16*8(sp)\r |
| | 115 | movaps %xmm9, 16*9(sp)\r |
| | 116 | movaps %xmm10, 16*10(sp)\r |
| | 117 | movaps %xmm11, 16*11(sp)\r |
| | 118 | movaps %xmm12, 16*12(sp)\r |
| | 119 | movaps %xmm13, 16*13(sp)\r |
| | 120 | movaps %xmm14, 16*14(sp)\r |
| | 121 | movaps %xmm15, 16*15(sp)\r |
| | 122 | #endif // __x86_64__\r |
| | 123 | \r |
| | 124 | #endif // KERNEL\r |
| | 125 | \r |
| | 126 | #define iv %xmm0\r |
| | 127 | \r |
| | 128 | #ifdef __i386__\r |
| | 129 | \r |
| | 130 | mov 12(%ebp), %eax // in_iv\r |
| | 131 | mov 24(%ebp), %edx // ctx\r |
| | 132 | movups (%eax), iv // iv = in_iv \r |
| | 133 | mov 8(%ebp), %ebx // ibuf\r |
| | 134 | mov 16(%ebp), %ecx // num_blk\r |
| | 135 | mov 20(%ebp), %edi // obuf\r |
| | 136 | \r |
| | 137 | #define ibuf %ebx\r |
| | 138 | #define obuf %edi\r |
| | 139 | #define num_blk %ecx \r |
| | 140 | #define ctx %edx\r |
| | 141 | \r |
| | 142 | #else\r |
| | 143 | \r |
| | 144 | mov %rdi, %rbx // ibuf\r |
| | 145 | movups (%rsi), iv // iv = in_iv\r |
| | 146 | mov %rdx, %r13 // num_blk\r |
| | 147 | mov %rcx, %r14 // obuf\r |
| | 148 | mov %r8, %r15 // ctx \r |
| | 149 | \r |
| | 150 | #define ibuf %rbx\r |
| | 151 | #define num_blk %r13d\r |
| | 152 | #define obuf %r14 \r |
| | 153 | #define ctx %r15\r |
| | 154 | \r |
| | 155 | #endif\r |
| | 156 | \r |
| | 157 | mov 240(ctx), %eax // aes length\r |
| | 158 | cmp $160, %eax // aes-128 encrypt ?\r |
| | 159 | je L_encrypt_128\r |
| | 160 | cmp $192, %eax // aes-192 encrypt ?\r |
| | 161 | je L_encrypt_192\r |
| | 162 | cmp $224, %eax // aes-256 encrypt ?\r |
| | 163 | je L_encrypt_256\r |
| | 164 | mov $-1, %eax // return error\r |
| | 165 | jmp L_error \r |
| | 166 | \r |
| | 167 | //\r |
| | 168 | // aes-128 encrypt_cbc operation, up to L_HW_cbc_done\r |
| | 169 | //\r |
| | 170 | \r |
| | 171 | L_encrypt_128:\r |
| | 172 | \r |
| | 173 | cmp $1, num_blk // check number of block\r |
| | 174 | jl L_HW_cbc_done // should it be less than 1, nothing to do\r |
| | 175 | \r |
| | 176 | movups (ctx), %xmm2 // key0\r |
| | 177 | movups 16(ctx), %xmm3 // key1\r |
| | 178 | movups 32(ctx), %xmm4 // key2\r |
| | 179 | movups 48(ctx), %xmm5 // key3\r |
| | 180 | movups 64(ctx), %xmm6 // key4\r |
| | 181 | movups 80(ctx), %xmm7 // key5\r |
| | 182 | #if defined __x86_64__\r |
| | 183 | movups 96(ctx), %xmm8 // key6\r |
| | 184 | movups 112(ctx), %xmm9 // key7\r |
| | 185 | movups 128(ctx), %xmm10 // key8\r |
| | 186 | movups 144(ctx), %xmm11 // key9\r |
| | 187 | movups 160(ctx), %xmm12 // keyA\r |
| | 188 | #endif\r |
| | 189 | \r |
| | 190 | // while (num_blk--) {\r |
| | 191 | // *iv ^= *ibuf++;\r |
| | 192 | // aes_encrypt(iv, iv, ctx);\r |
| | 193 | // *obuf++ = *iv;\r |
| | 194 | // }\r |
| | 195 | 0:\r |
| | 196 | movups (ibuf), %xmm1 // *ibuf\r |
| | 197 | pxor %xmm2, iv // 1st instruction inside aes_encrypt\r |
| | 198 | pxor %xmm1, iv // *iv ^= *ibuf\r |
| | 199 | \r |
| | 200 | // finishing up the rest of aes_encrypt\r |
| | 201 | aesenc %xmm3, iv\r |
| | 202 | aesenc %xmm4, iv\r |
| | 203 | aesenc %xmm5, iv\r |
| | 204 | aesenc %xmm6, iv\r |
| | 205 | aesenc %xmm7, iv\r |
| | 206 | #if defined __x86_64__\r |
| | 207 | aesenc %xmm8, iv\r |
| | 208 | aesenc %xmm9, iv\r |
| | 209 | aesenc %xmm10, iv\r |
| | 210 | aesenc %xmm11, iv\r |
| | 211 | aesenclast %xmm12, iv\r |
| | 212 | #else\r |
| | 213 | movups 96(ctx), %xmm1 // key6\r |
| | 214 | aesenc %xmm1, iv\r |
| | 215 | movups 112(ctx), %xmm1 // key7\r |
| | 216 | aesenc %xmm1, iv\r |
| | 217 | movups 128(ctx), %xmm1 // key8\r |
| | 218 | aesenc %xmm1, iv\r |
| | 219 | movups 144(ctx), %xmm1 // key9\r |
| | 220 | aesenc %xmm1, iv\r |
| | 221 | movups 160(ctx), %xmm1 // keyA\r |
| | 222 | aesenclast %xmm1, iv\r |
| | 223 | #endif\r |
| | 224 | \r |
| | 225 | movups iv, (obuf) // *obuf = *iv;\r |
| | 226 | add $16, obuf // obuf++;\r |
| | 227 | add $16, ibuf // ibuf++;\r |
| | 228 | sub $1, num_blk // num_blk --\r |
| | 229 | jg 0b // if num_blk > 0, repeat the loop\r |
| | 230 | \r |
| | 231 | // the following will be branched to from all other cases (encrypt/decrypt 128/192/256)\r |
| | 232 | \r |
| | 233 | L_HW_cbc_done:\r |
| | 234 | \r |
| | 235 | xor %eax, %eax // to return CRYPT_OK\r |
| | 236 | \r |
| | 237 | L_error:\r |
| | 238 | \r |
| | 239 | // if kernel, restore xmm registers\r |
| | 240 | #ifdef KERNEL \r |
| | 241 | movaps 0(sp), %xmm0\r |
| | 242 | movaps 16(sp), %xmm1\r |
| | 243 | movaps 32(sp), %xmm2\r |
| | 244 | movaps 48(sp), %xmm3\r |
| | 245 | movaps 64(sp), %xmm4\r |
| | 246 | movaps 80(sp), %xmm5\r |
| | 247 | movaps 96(sp), %xmm6\r |
| | 248 | movaps 112(sp), %xmm7\r |
| | 249 | #if defined __x86_64__\r |
| | 250 | movaps 16*8(sp), %xmm8\r |
| | 251 | movaps 16*9(sp), %xmm9\r |
| | 252 | movaps 16*10(sp), %xmm10\r |
| | 253 | movaps 16*11(sp), %xmm11\r |
| | 254 | movaps 16*12(sp), %xmm12\r |
| | 255 | movaps 16*13(sp), %xmm13\r |
| | 256 | movaps 16*14(sp), %xmm14\r |
| | 257 | movaps 16*15(sp), %xmm15\r |
| | 258 | #endif // __x86_64__\r |
| | 259 | #endif // KERNEL\r |
| | 260 | \r |
| | 261 | // release used stack memory, restore used callee-saved registers, and return \r |
| | 262 | #if defined __i386__\r |
| | 263 | #ifdef KERNEL\r |
| | 264 | add $(8*16), %esp\r |
| | 265 | #endif\r |
| | 266 | pop %edi\r |
| | 267 | pop %ebx\r |
| | 268 | #else\r |
| | 269 | #ifdef KERNEL\r |
| | 270 | add $(16*16), %rsp \r |
| | 271 | #endif\r |
| | 272 | pop %r15\r |
| | 273 | pop %r14\r |
| | 274 | pop %r13\r |
| | 275 | pop %rbx\r |
| | 276 | #endif\r |
| | 277 | leave\r |
| | 278 | ret\r |
| | 279 | \r |
| | 280 | //\r |
| | 281 | // aes-192 encrypt_cbc operation, after completion, branch to L_HW_cbc_done\r |
| | 282 | //\r |
| | 283 | \r |
| | 284 | L_encrypt_192:\r |
| | 285 | \r |
| | 286 | cmp $1, num_blk // check number of block\r |
| | 287 | jl L_HW_cbc_done // should it be less than 1, nothing to do\r |
| | 288 | \r |
| | 289 | movups (ctx), %xmm2 // key0\r |
| | 290 | movups 16(ctx), %xmm3 // key1\r |
| | 291 | movups 32(ctx), %xmm4 // key2\r |
| | 292 | movups 48(ctx), %xmm5 // key3\r |
| | 293 | movups 64(ctx), %xmm6 // key4\r |
| | 294 | movups 80(ctx), %xmm7 // key5\r |
| | 295 | #if defined __x86_64__\r |
| | 296 | movups 96(ctx), %xmm8 // key6\r |
| | 297 | movups 112(ctx), %xmm9 // key7\r |
| | 298 | movups 128(ctx), %xmm10 // key8\r |
| | 299 | movups 144(ctx), %xmm11 // key9\r |
| | 300 | movups 160(ctx), %xmm12 // keyA\r |
| | 301 | movups 176(ctx), %xmm13 // keyB\r |
| | 302 | movups 192(ctx), %xmm14 // keyC\r |
| | 303 | #endif\r |
| | 304 | \r |
| | 305 | // while (num_blk--) {\r |
| | 306 | // *iv ^= *ibuf++;\r |
| | 307 | // aes_encrypt(iv, iv, ctx);\r |
| | 308 | // *obuf++ = *iv;\r |
| | 309 | // }\r |
| | 310 | 0:\r |
| | 311 | movups (ibuf), %xmm1 // *ibuf\r |
| | 312 | pxor %xmm1, iv // *iv ^= ibuf\r |
| | 313 | \r |
| | 314 | // aes_encrypt(iv, iv, ctx);\r |
| | 315 | \r |
| | 316 | pxor %xmm2, iv\r |
| | 317 | aesenc %xmm3, iv\r |
| | 318 | aesenc %xmm4, iv\r |
| | 319 | aesenc %xmm5, iv\r |
| | 320 | aesenc %xmm6, iv\r |
| | 321 | aesenc %xmm7, iv\r |
| | 322 | #if defined __x86_64__\r |
| | 323 | aesenc %xmm8, iv\r |
| | 324 | aesenc %xmm9, iv\r |
| | 325 | aesenc %xmm10, iv\r |
| | 326 | aesenc %xmm11, iv\r |
| | 327 | aesenc %xmm12, iv\r |
| | 328 | aesenc %xmm13, iv\r |
| | 329 | aesenclast %xmm14, iv\r |
| | 330 | #else\r |
| | 331 | movups 96(ctx), %xmm1\r |
| | 332 | aesenc %xmm1, iv\r |
| | 333 | movups 112(ctx), %xmm1\r |
| | 334 | aesenc %xmm1, iv\r |
| | 335 | movups 128(ctx), %xmm1\r |
| | 336 | aesenc %xmm1, iv\r |
| | 337 | movups 144(ctx), %xmm1\r |
| | 338 | aesenc %xmm1, iv\r |
| | 339 | movups 160(ctx), %xmm1\r |
| | 340 | aesenc %xmm1, iv\r |
| | 341 | movups 176(ctx), %xmm1\r |
| | 342 | aesenc %xmm1, iv\r |
| | 343 | movups 192(ctx), %xmm1\r |
| | 344 | aesenclast %xmm1, iv\r |
| | 345 | #endif\r |
| | 346 | \r |
| | 347 | movups iv, (obuf) // *obuf = *iv;\r |
| | 348 | add $16, ibuf // ibuf++\r |
| | 349 | add $16, obuf // obuf++\r |
| | 350 | \r |
| | 351 | sub $1, num_blk // num_blk --\r |
| | 352 | jg 0b // if num_blk > 0, repeat the loop\r |
| | 353 | \r |
| | 354 | jmp L_HW_cbc_done // share with the common exit code\r |
| | 355 | \r |
| | 356 | //\r |
| | 357 | // aes-256 encrypt_cbc operation, after completion, branch to L_HW_cbc_done\r |
| | 358 | //\r |
| | 359 | \r |
| | 360 | L_encrypt_256:\r |
| | 361 | \r |
| | 362 | cmp $1, num_blk // check number of block\r |
| | 363 | jl L_HW_cbc_done // should it be less than 1, nothing to do\r |
| | 364 | \r |
| | 365 | movups (ctx), %xmm2 // key0\r |
| | 366 | movups 16(ctx), %xmm3 // key1\r |
| | 367 | movups 32(ctx), %xmm4 // key2\r |
| | 368 | movups 48(ctx), %xmm5 // key3\r |
| | 369 | movups 64(ctx), %xmm6 // key4\r |
| | 370 | movups 80(ctx), %xmm7 // key5\r |
| | 371 | #if defined __x86_64__\r |
| | 372 | movups 96(ctx), %xmm8 // key6\r |
| | 373 | movups 112(ctx), %xmm9 // key7\r |
| | 374 | movups 128(ctx), %xmm10 // key8\r |
| | 375 | movups 144(ctx), %xmm11 // key9\r |
| | 376 | movups 160(ctx), %xmm12 // keyA\r |
| | 377 | movups 176(ctx), %xmm13 // keyB\r |
| | 378 | movups 192(ctx), %xmm14 // keyC\r |
| | 379 | movups 208(ctx), %xmm15 // keyD\r |
| | 380 | // movups 224(ctx), %xmm1 // keyE\r |
| | 381 | #endif\r |
| | 382 | \r |
| | 383 | // while (num_blk--) {\r |
| | 384 | // *iv ^= *ibuf++;\r |
| | 385 | // aes_encrypt(iv, iv, ctx);\r |
| | 386 | // *obuf++ = *iv;\r |
| | 387 | // }\r |
| | 388 | 0:\r |
| | 389 | movups (ibuf), %xmm1 // *ibuf\r |
| | 390 | pxor %xmm1, iv // *iv ^= ibuf\r |
| | 391 | \r |
| | 392 | // aes_encrypt(iv, iv, ctx);\r |
| | 393 | pxor %xmm2, iv\r |
| | 394 | aesenc %xmm3, iv\r |
| | 395 | aesenc %xmm4, iv\r |
| | 396 | aesenc %xmm5, iv\r |
| | 397 | aesenc %xmm6, iv\r |
| | 398 | aesenc %xmm7, iv\r |
| | 399 | #if defined __x86_64__\r |
| | 400 | movups 224(ctx), %xmm1 // keyE\r |
| | 401 | aesenc %xmm8, iv\r |
| | 402 | aesenc %xmm9, iv\r |
| | 403 | aesenc %xmm10, iv\r |
| | 404 | aesenc %xmm11, iv\r |
| | 405 | aesenc %xmm12, iv\r |
| | 406 | aesenc %xmm13, iv\r |
| | 407 | aesenc %xmm14, iv\r |
| | 408 | aesenc %xmm15, iv\r |
| | 409 | aesenclast %xmm1, iv\r |
| | 410 | #else\r |
| | 411 | movups 96(ctx), %xmm1 // key6\r |
| | 412 | aesenc %xmm1, iv\r |
| | 413 | movups 112(ctx), %xmm1 // key7\r |
| | 414 | aesenc %xmm1, iv\r |
| | 415 | movups 128(ctx), %xmm1 // key8\r |
| | 416 | aesenc %xmm1, iv\r |
| | 417 | movups 144(ctx), %xmm1 // key9\r |
| | 418 | aesenc %xmm1, iv\r |
| | 419 | movups 160(ctx), %xmm1 // keyA\r |
| | 420 | aesenc %xmm1, iv\r |
| | 421 | movups 176(ctx), %xmm1 // keyB\r |
| | 422 | aesenc %xmm1, iv\r |
| | 423 | movups 192(ctx), %xmm1 // keyC\r |
| | 424 | aesenc %xmm1, iv\r |
| | 425 | movups 208(ctx), %xmm1 // keyD\r |
| | 426 | aesenc %xmm1, iv\r |
| | 427 | movups 224(ctx), %xmm1 // keyE\r |
| | 428 | aesenclast %xmm1, iv\r |
| | 429 | #endif\r |
| | 430 | \r |
| | 431 | movups iv, (obuf) // *obuf = *iv;\r |
| | 432 | add $16, ibuf // ibuf++\r |
| | 433 | add $16, obuf // obuf++\r |
| | 434 | \r |
| | 435 | sub $1, num_blk // num_blk --\r |
| | 436 | jg 0b // if num_blk > 0, repeat the loop\r |
| | 437 | \r |
| | 438 | jmp L_HW_cbc_done // share with the common exit code\r |
| | 439 | \r |
| | 440 | \r |
| | 441 | \r |
| | 442 | //\r |
| | 443 | // --------- END of aes_encrypt_cbc_hw -------------------\r |
| | 444 | //\r |
| | 445 | \r |
| | 446 | \r |
| | 447 | /* ---------------------------------------------------------------------------------------------------------------- \r |
| | 448 | \r |
| | 449 | aes_decrypt_cbc function (see aes_modes.c or aes_modes_asm.s) :\r |
| | 450 | \r |
| | 451 | For simplicity, I am assuming all variables are in 128-bit data type.\r |
| | 452 | \r |
| | 453 | aes_rval aes_decrypt_cbc(const __m128 *ibuf, __m128 *iv, int num_blk, __m128 *obuf, const aes_decrypt_ctx *ctx)\r |
| | 454 | {\r |
| | 455 | while(num_blk--) {\r |
| | 456 | aes_decrypt(ibuf, obuf, ctx);\r |
| | 457 | *obuf++ ^= *iv;\r |
| | 458 | *iv = *ibuf++;\r |
| | 459 | }\r |
| | 460 | return 0;\r |
| | 461 | }\r |
| | 462 | \r |
| | 463 | The following is an implementation of this function using Intel AESNI.\r |
| | 464 | This function _aes_decrypt_cbc_hw SHOULD NOT be called directly. \r |
| | 465 | Developer should still call _aes_decrypt_cbc (in aes_modes_asm.s) which will poll cpu_capabilities and branch\r |
| | 466 | to this aesni-based function should it detecs that aesni is available.\r |
| | 467 | Blindly call this function SURELY will cause a CRASH on systems with no aesni support. \r |
| | 468 | \r |
| | 469 | Note that the decryption operation is not related over blocks.\r |
| | 470 | This gives opportunity of arranging aes_decrypt operations in parallel to speed up code.\r |
| | 471 | This is equivalent to what has been described in the Intel AES Instruction Set White Paper (Rev. 2.0 page 53-55)\r |
| | 472 | The following assembly code exploits this idea to achieve ~ 1.4 speed up in aes_decrypt_cbc.\r |
| | 473 | \r |
| | 474 | Example C code for packing 4 blocks in an iteration is shown as follows:\r |
| | 475 | \r |
| | 476 | while ((num_blk-=4)>=0) {\r |
| | 477 | \r |
| | 478 | // the following 4 functions can be interleaved to exploit parallelism\r |
| | 479 | aes_decrypt(ibuf, obuf, ctx);\r |
| | 480 | aes_decrypt(ibuf+1, obuf+1, ctx);\r |
| | 481 | aes_decrypt(ibuf+2, obuf+2, ctx);\r |
| | 482 | aes_decrypt(ibuf+3, obuf+3, ctx);\r |
| | 483 | \r |
| | 484 | obuf[0] ^= *iv; obuf[1] ^= ibuf[1]; obuf[2] ^= ibuf[1]; obuf[3] ^= ibuf[2];\r |
| | 485 | *iv = ibuf[3]; ibuf += 4; obuf += 4;\r |
| | 486 | }\r |
| | 487 | num_blk+=4;\r |
| | 488 | \r |
| | 489 | ----------------------------------------------------------------------------------------------------------------*/\r |
| | 490 | \r |
| | 491 | .text\r |
| | 492 | .align 4,0x90\r |
| | 493 | .globl _aes_decrypt_cbc_hw\r |
| | 494 | _aes_decrypt_cbc_hw:\r |
| | 495 | \r |
| | 496 | // push/save registers for local use\r |
| | 497 | #if defined __i386__\r |
| | 498 | \r |
| | 499 | push %ebp\r |
| | 500 | movl %esp, %ebp\r |
| | 501 | push %ebx // ibuf\r |
| | 502 | push %edi // obuf\r |
| | 503 | \r |
| | 504 | #define sp %esp\r |
| | 505 | \r |
| | 506 | #else // __x86_64__\r |
| | 507 | \r |
| | 508 | push %rbp\r |
| | 509 | mov %rsp, %rbp\r |
| | 510 | push %rbx\r |
| | 511 | push %r13\r |
| | 512 | push %r14\r |
| | 513 | push %r15\r |
| | 514 | \r |
| | 515 | #define sp %rsp\r |
| | 516 | \r |
| | 517 | #endif\r |
| | 518 | \r |
| | 519 | \r |
| | 520 | // if kernel, allocate stack space to save xmm registers\r |
| | 521 | #ifdef KERNEL\r |
| | 522 | #if defined __i386__\r |
| | 523 | sub $(8*16), %esp\r |
| | 524 | #else\r |
| | 525 | sub $(16*16), %rsp\r |
| | 526 | #endif\r |
| | 527 | movaps %xmm0, (sp)\r |
| | 528 | movaps %xmm1, 16(sp)\r |
| | 529 | movaps %xmm2, 32(sp)\r |
| | 530 | movaps %xmm3, 48(sp)\r |
| | 531 | movaps %xmm4, 64(sp)\r |
| | 532 | movaps %xmm5, 80(sp)\r |
| | 533 | movaps %xmm6, 96(sp)\r |
| | 534 | movaps %xmm7, 112(sp)\r |
| | 535 | #if defined __x86_64__\r |
| | 536 | movaps %xmm8, 16*8(sp)\r |
| | 537 | movaps %xmm9, 16*9(sp)\r |
| | 538 | movaps %xmm10, 16*10(sp)\r |
| | 539 | movaps %xmm11, 16*11(sp)\r |
| | 540 | movaps %xmm12, 16*12(sp)\r |
| | 541 | movaps %xmm13, 16*13(sp)\r |
| | 542 | movaps %xmm14, 16*14(sp)\r |
| | 543 | movaps %xmm15, 16*15(sp)\r |
| | 544 | #endif // __x86_64__\r |
| | 545 | #endif\r |
| | 546 | \r |
| | 547 | #undef iv\r |
| | 548 | #define iv %xmm0\r |
| | 549 | \r |
| | 550 | #if defined __i386__\r |
| | 551 | mov 12(%ebp), %eax // in_iv\r |
| | 552 | mov 24(%ebp), %edx // ctx\r |
| | 553 | movups (%eax), iv // iv = in_iv \r |
| | 554 | mov 8(%ebp), %ebx // ibuf\r |
| | 555 | mov 16(%ebp), %ecx // num_blk\r |
| | 556 | mov 20(%ebp), %edi // obuf\r |
| | 557 | \r |
| | 558 | #define ibuf %ebx\r |
| | 559 | #define obuf %edi\r |
| | 560 | #define num_blk %ecx \r |
| | 561 | #define ctx %edx\r |
| | 562 | \r |
| | 563 | #else // __x86_64__, rdi/rsi/rdx/rcx/r8\r |
| | 564 | \r |
| | 565 | mov %rdi, %rbx // ibuf\r |
| | 566 | movups (%rsi), iv // iv = in_iv\r |
| | 567 | mov %rdx, %r13 // num_blk\r |
| | 568 | mov %rcx, %r14 // obuf\r |
| | 569 | mov %r8, %r15 // ctx \r |
| | 570 | \r |
| | 571 | #define ibuf %rbx\r |
| | 572 | #define num_blk %r13d\r |
| | 573 | #define obuf %r14 \r |
| | 574 | #define ctx %r15\r |
| | 575 | \r |
| | 576 | #endif\r |
| | 577 | \r |
| | 578 | mov 240(ctx), %eax // aes length\r |
| | 579 | cmp $160, %eax // aes-128 decrypt\r |
| | 580 | je L_decrypt_128\r |
| | 581 | cmp $192, %eax // aes-192 decrypt\r |
| | 582 | je L_decrypt_192\r |
| | 583 | cmp $224, %eax // aes-256 decrypt\r |
| | 584 | je L_decrypt_256\r |
| | 585 | \r |
| | 586 | mov $-1, %eax // wrong aes length, to return -1\r |
| | 587 | jmp L_error // early exit due to wrong aes length\r |
| | 588 | \r |
| | 589 | \r |
| | 590 | //\r |
| | 591 | // aes-128 decrypt_cbc operation, after completion, branch to L_HW_cbc_done\r |
| | 592 | //\r |
| | 593 | \r |
| | 594 | L_decrypt_128:\r |
| | 595 | \r |
| | 596 | cmp $1, num_blk\r |
| | 597 | jl L_HW_cbc_done // if num_blk < 1, early return\r |
| | 598 | \r |
| | 599 | // aes-128 decrypt expanded keys\r |
| | 600 | movups 160(ctx), %xmm3\r |
| | 601 | movups 144(ctx), %xmm4\r |
| | 602 | movups 128(ctx), %xmm5\r |
| | 603 | movups 112(ctx), %xmm6\r |
| | 604 | movups 96(ctx), %xmm7\r |
| | 605 | #if defined __x86_64__\r |
| | 606 | movups 80(ctx), %xmm8\r |
| | 607 | movups 64(ctx), %xmm9\r |
| | 608 | movups 48(ctx), %xmm10\r |
| | 609 | movups 32(ctx), %xmm11\r |
| | 610 | movups 16(ctx), %xmm12\r |
| | 611 | movups 0(ctx), %xmm13\r |
| | 612 | #endif\r |
| | 613 | \r |
| | 614 | // performs 4 block decryption in an iteration to exploit decrypt in parallel\r |
| | 615 | \r |
| | 616 | // while ((num_blk-=4)>=0) {\r |
| | 617 | // aes_decrypt(ibuf, obuf, ctx);\r |
| | 618 | // aes_decrypt(ibuf+1, obuf+1, ctx);\r |
| | 619 | // aes_decrypt(ibuf+2, obuf+2, ctx);\r |
| | 620 | // aes_decrypt(ibuf+3, obuf+3, ctx);\r |
| | 621 | // obuf[0] ^= *iv; obuf[1] ^= ibuf[1]; obuf[2] ^= ibuf[1]; obuf[3] ^= ibuf[2];\r |
| | 622 | // *iv = ibuf[3]; ibuf += 4; obuf += 4;\r |
| | 623 | // }\r |
| | 624 | \r |
| | 625 | sub $4, num_blk // pre decrement num_blk by 4\r |
| | 626 | jl 9f // if num_blk < 4, skip the per-4-blocks processing code\r |
| | 627 | \r |
| | 628 | 0:\r |
| | 629 | \r |
| | 630 | \r |
| | 631 | #if defined __x86_64__\r |
| | 632 | \r |
| | 633 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r |
| | 634 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r |
| | 635 | movups 32(ibuf), %xmm14 // tmp = 3rd ibuf\r |
| | 636 | movups 48(ibuf), %xmm15 // tmp = 4th ibuf\r |
| | 637 | \r |
| | 638 | // for x86_64, the expanded keys are already stored in xmm3-xmm13\r |
| | 639 | \r |
| | 640 | // aes-128 decrypt round 0 per 4 blocks\r |
| | 641 | pxor %xmm3, %xmm1\r |
| | 642 | pxor %xmm3, %xmm2\r |
| | 643 | pxor %xmm3, %xmm14\r |
| | 644 | pxor %xmm3, %xmm15\r |
| | 645 | \r |
| | 646 | // aes-128 decrypt round 1 per 4 blocks\r |
| | 647 | aesdec %xmm4, %xmm1\r |
| | 648 | aesdec %xmm4, %xmm2\r |
| | 649 | aesdec %xmm4, %xmm14\r |
| | 650 | aesdec %xmm4, %xmm15\r |
| | 651 | \r |
| | 652 | // aes-128 decrypt round 2 per 4 blocks\r |
| | 653 | aesdec %xmm5, %xmm1\r |
| | 654 | aesdec %xmm5, %xmm2\r |
| | 655 | aesdec %xmm5, %xmm14\r |
| | 656 | aesdec %xmm5, %xmm15\r |
| | 657 | \r |
| | 658 | // aes-128 decrypt round 3 per 4 blocks\r |
| | 659 | aesdec %xmm6, %xmm1\r |
| | 660 | aesdec %xmm6, %xmm2\r |
| | 661 | aesdec %xmm6, %xmm14\r |
| | 662 | aesdec %xmm6, %xmm15\r |
| | 663 | \r |
| | 664 | // aes-128 decrypt round 4 per 4 blocks\r |
| | 665 | aesdec %xmm7, %xmm1\r |
| | 666 | aesdec %xmm7, %xmm2\r |
| | 667 | aesdec %xmm7, %xmm14\r |
| | 668 | aesdec %xmm7, %xmm15\r |
| | 669 | \r |
| | 670 | // aes-128 decrypt round 5 per 4 blocks\r |
| | 671 | aesdec %xmm8, %xmm1\r |
| | 672 | aesdec %xmm8, %xmm2\r |
| | 673 | aesdec %xmm8, %xmm14\r |
| | 674 | aesdec %xmm8, %xmm15\r |
| | 675 | \r |
| | 676 | // aes-128 decrypt round 6 per 4 blocks\r |
| | 677 | aesdec %xmm9, %xmm1\r |
| | 678 | aesdec %xmm9, %xmm2\r |
| | 679 | aesdec %xmm9, %xmm14\r |
| | 680 | aesdec %xmm9, %xmm15\r |
| | 681 | \r |
| | 682 | // aes-128 decrypt round 7 per 4 blocks\r |
| | 683 | aesdec %xmm10, %xmm1\r |
| | 684 | aesdec %xmm10, %xmm2\r |
| | 685 | aesdec %xmm10, %xmm14\r |
| | 686 | aesdec %xmm10, %xmm15\r |
| | 687 | \r |
| | 688 | // aes-128 decrypt round 8 per 4 blocks\r |
| | 689 | aesdec %xmm11, %xmm1\r |
| | 690 | aesdec %xmm11, %xmm2\r |
| | 691 | aesdec %xmm11, %xmm14\r |
| | 692 | aesdec %xmm11, %xmm15\r |
| | 693 | \r |
| | 694 | // aes-128 decrypt round 9 per 4 blocks\r |
| | 695 | aesdec %xmm12, %xmm1\r |
| | 696 | aesdec %xmm12, %xmm2\r |
| | 697 | aesdec %xmm12, %xmm14\r |
| | 698 | aesdec %xmm12, %xmm15\r |
| | 699 | \r |
| | 700 | // aes-128 decrypt round 10 (last) per 4 blocks\r |
| | 701 | aesdeclast %xmm13, %xmm1\r |
| | 702 | aesdeclast %xmm13, %xmm2\r |
| | 703 | aesdeclast %xmm13, %xmm14\r |
| | 704 | aesdeclast %xmm13, %xmm15\r |
| | 705 | \r |
| | 706 | pxor iv, %xmm1 // obuf[0] ^= *iv; \r |
| | 707 | movups (ibuf), iv // ibuf[0]\r |
| | 708 | pxor iv, %xmm2 // obuf[1] ^= ibuf[0]; \r |
| | 709 | movups 16(ibuf), iv // ibuf[1]\r |
| | 710 | pxor iv, %xmm14 // obuf[2] ^= ibuf[1]; \r |
| | 711 | movups 32(ibuf), iv // ibuf[2] \r |
| | 712 | pxor iv, %xmm15 // obuf[3] ^= obuf[2]; \r |
| | 713 | movups 48(ibuf), iv // *iv = ibuf[3]\r |
| | 714 | \r |
| | 715 | movups %xmm1, (obuf) // write 1st obuf\r |
| | 716 | movups %xmm2, 16(obuf) // write 2nd obuf\r |
| | 717 | movups %xmm14, 32(obuf) // write 3rd obuf\r |
| | 718 | movups %xmm15, 48(obuf) // write 4th obuf\r |
| | 719 | \r |
| | 720 | \r |
| | 721 | #else\r |
| | 722 | \r |
| | 723 | // aes_decrypt_cbc per 4 blocks using aes-128 for i386\r |
| | 724 | // xmm1/xmm2/xmm4/xmm5 used for obuf per block\r |
| | 725 | // xmm3 = key0\r |
| | 726 | // xmm0 = iv\r |
| | 727 | // xmm6/xmm7 dynamically load with other expanded keys\r |
| | 728 | \r |
| | 729 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r |
| | 730 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r |
| | 731 | movups 32(ibuf), %xmm4 // tmp = 3rd ibuf\r |
| | 732 | movups 48(ibuf), %xmm5 // tmp = 4th ibuf\r |
| | 733 | \r |
| | 734 | // aes_decrypt\r |
| | 735 | // for i386, sequentially load expanded keys into xmm6/xmm7\r |
| | 736 | \r |
| | 737 | movups 144(ctx), %xmm6 // key1\r |
| | 738 | \r |
| | 739 | // aes-128 decrypt round 0 per 4 blocks\r |
| | 740 | pxor %xmm3, %xmm1\r |
| | 741 | pxor %xmm3, %xmm2\r |
| | 742 | pxor %xmm3, %xmm4\r |
| | 743 | pxor %xmm3, %xmm5\r |
| | 744 | \r |
| | 745 | movups 128(ctx), %xmm7 // key2\r |
| | 746 | \r |
| | 747 | // aes-128 decrypt round 1 per 4 blocks\r |
| | 748 | aesdec %xmm6, %xmm1\r |
| | 749 | aesdec %xmm6, %xmm2\r |
| | 750 | aesdec %xmm6, %xmm4\r |
| | 751 | aesdec %xmm6, %xmm5\r |
| | 752 | \r |
| | 753 | movups 112(ctx), %xmm6 // key3\r |
| | 754 | \r |
| | 755 | // aes-128 decrypt round 2 per 4 blocks\r |
| | 756 | aesdec %xmm7, %xmm1\r |
| | 757 | aesdec %xmm7, %xmm2\r |
| | 758 | aesdec %xmm7, %xmm4\r |
| | 759 | aesdec %xmm7, %xmm5\r |
| | 760 | \r |
| | 761 | movups 96(ctx), %xmm7 // key4\r |
| | 762 | \r |
| | 763 | // aes-128 decrypt round 3 per 4 blocks\r |
| | 764 | aesdec %xmm6, %xmm1\r |
| | 765 | aesdec %xmm6, %xmm2\r |
| | 766 | aesdec %xmm6, %xmm4\r |
| | 767 | aesdec %xmm6, %xmm5\r |
| | 768 | \r |
| | 769 | movups 80(ctx), %xmm6 // key5\r |
| | 770 | \r |
| | 771 | // aes-128 decrypt round 4 per 4 blocks\r |
| | 772 | aesdec %xmm7, %xmm1\r |
| | 773 | aesdec %xmm7, %xmm2\r |
| | 774 | aesdec %xmm7, %xmm4\r |
| | 775 | aesdec %xmm7, %xmm5\r |
| | 776 | \r |
| | 777 | movups 64(ctx), %xmm7 // key6\r |
| | 778 | \r |
| | 779 | // aes-128 decrypt round 5 per 4 blocks\r |
| | 780 | aesdec %xmm6, %xmm1\r |
| | 781 | aesdec %xmm6, %xmm2\r |
| | 782 | aesdec %xmm6, %xmm4\r |
| | 783 | aesdec %xmm6, %xmm5\r |
| | 784 | \r |
| | 785 | movups 48(ctx), %xmm6 // key7\r |
| | 786 | \r |
| | 787 | // aes-128 decrypt round 6 per 4 blocks\r |
| | 788 | aesdec %xmm7, %xmm1\r |
| | 789 | aesdec %xmm7, %xmm2\r |
| | 790 | aesdec %xmm7, %xmm4\r |
| | 791 | aesdec %xmm7, %xmm5\r |
| | 792 | \r |
| | 793 | movups 32(ctx), %xmm7 // key8\r |
| | 794 | \r |
| | 795 | // aes-128 decrypt round 7 per 4 blocks\r |
| | 796 | aesdec %xmm6, %xmm1\r |
| | 797 | aesdec %xmm6, %xmm2\r |
| | 798 | aesdec %xmm6, %xmm4\r |
| | 799 | aesdec %xmm6, %xmm5\r |
| | 800 | \r |
| | 801 | movups 16(ctx), %xmm6 // key9\r |
| | 802 | \r |
| | 803 | // aes-128 decrypt round 8 per 4 blocks\r |
| | 804 | aesdec %xmm7, %xmm1\r |
| | 805 | aesdec %xmm7, %xmm2\r |
| | 806 | aesdec %xmm7, %xmm4\r |
| | 807 | aesdec %xmm7, %xmm5\r |
| | 808 | \r |
| | 809 | movups 0(ctx), %xmm7 // keyA\r |
| | 810 | \r |
| | 811 | // aes-128 decrypt round 9 per 4 blocks\r |
| | 812 | aesdec %xmm6, %xmm1\r |
| | 813 | aesdec %xmm6, %xmm2\r |
| | 814 | aesdec %xmm6, %xmm4\r |
| | 815 | aesdec %xmm6, %xmm5\r |
| | 816 | \r |
| | 817 | // aes-128 decrypt round 10 (last) per 4 blocks\r |
| | 818 | aesdeclast %xmm7, %xmm1\r |
| | 819 | aesdeclast %xmm7, %xmm2\r |
| | 820 | aesdeclast %xmm7, %xmm4\r |
| | 821 | aesdeclast %xmm7, %xmm5\r |
| | 822 | \r |
| | 823 | pxor iv, %xmm1 // 1st obuf ^= iv; \r |
| | 824 | movups (ibuf), iv // 1st memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 825 | pxor iv, %xmm2 // 2nd obuf ^= iv; \r |
| | 826 | movups 16(ibuf), iv // 2nd memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 827 | pxor iv, %xmm4 // 3rd obuf ^= iv; \r |
| | 828 | movups 32(ibuf), iv // 3rd memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 829 | pxor iv, %xmm5 // 4th obuf ^= iv; \r |
| | 830 | movups 48(ibuf), iv // 4th memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 831 | \r |
| | 832 | movups %xmm1, (obuf) // write 1st obuf\r |
| | 833 | movups %xmm2, 16(obuf) // write 2nd obuf\r |
| | 834 | movups %xmm4, 32(obuf) // write 3rd obuf\r |
| | 835 | movups %xmm5, 48(obuf) // write 4th obuf\r |
| | 836 | #endif\r |
| | 837 | \r |
| | 838 | add $64, ibuf // ibuf += 4; \r |
| | 839 | add $64, obuf // obuf += 4; \r |
| | 840 | \r |
| | 841 | sub $4, num_blk // num_blk -= 4\r |
| | 842 | jge 0b // if num_blk > 0, repeat the loop\r |
| | 843 | \r |
| | 844 | 9: add $4, num_blk // post incremtn num_blk by 4\r |
| | 845 | je L_HW_cbc_done // if num_blk == 0, no need for forthur processing code\r |
| | 846 | \r |
| | 847 | #if defined __i386__\r |
| | 848 | // updated as they might be needed as expanded keys in the remaining\r |
| | 849 | movups 144(ctx), %xmm4\r |
| | 850 | movups 128(ctx), %xmm5\r |
| | 851 | movups 112(ctx), %xmm6\r |
| | 852 | movups 96(ctx), %xmm7\r |
| | 853 | #endif\r |
| | 854 | \r |
| | 855 | test $2, num_blk // check whether num_blk has 2 blocks\r |
| | 856 | je 9f // if num_blk & 2 == 0, skip the per-pair processing code\r |
| | 857 | \r |
| | 858 | // do the remaining 2 blocks together\r |
| | 859 | \r |
| | 860 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r |
| | 861 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r |
| | 862 | \r |
| | 863 | // aes_decrypt\r |
| | 864 | pxor %xmm3, %xmm1\r |
| | 865 | pxor %xmm3, %xmm2\r |
| | 866 | aesdec %xmm4, %xmm1\r |
| | 867 | aesdec %xmm4, %xmm2\r |
| | 868 | aesdec %xmm5, %xmm1\r |
| | 869 | aesdec %xmm5, %xmm2\r |
| | 870 | aesdec %xmm6, %xmm1\r |
| | 871 | aesdec %xmm6, %xmm2\r |
| | 872 | #if defined __x86_64__\r |
| | 873 | aesdec %xmm7, %xmm1\r |
| | 874 | aesdec %xmm7, %xmm2\r |
| | 875 | aesdec %xmm8, %xmm1\r |
| | 876 | aesdec %xmm8, %xmm2\r |
| | 877 | aesdec %xmm9, %xmm1\r |
| | 878 | aesdec %xmm9, %xmm2\r |
| | 879 | aesdec %xmm10, %xmm1\r |
| | 880 | aesdec %xmm10, %xmm2\r |
| | 881 | aesdec %xmm11, %xmm1\r |
| | 882 | aesdec %xmm11, %xmm2\r |
| | 883 | aesdec %xmm12, %xmm1\r |
| | 884 | aesdec %xmm12, %xmm2\r |
| | 885 | aesdeclast %xmm13, %xmm1\r |
| | 886 | aesdeclast %xmm13, %xmm2\r |
| | 887 | #else\r |
| | 888 | movups 80(ctx), %xmm6\r |
| | 889 | aesdec %xmm7, %xmm1\r |
| | 890 | aesdec %xmm7, %xmm2\r |
| | 891 | movups 64(ctx), %xmm7\r |
| | 892 | aesdec %xmm6, %xmm1\r |
| | 893 | aesdec %xmm6, %xmm2\r |
| | 894 | movups 48(ctx), %xmm6\r |
| | 895 | aesdec %xmm7, %xmm1\r |
| | 896 | aesdec %xmm7, %xmm2\r |
| | 897 | movups 32(ctx), %xmm7\r |
| | 898 | aesdec %xmm6, %xmm1\r |
| | 899 | aesdec %xmm6, %xmm2\r |
| | 900 | movups 16(ctx), %xmm6\r |
| | 901 | aesdec %xmm7, %xmm1\r |
| | 902 | aesdec %xmm7, %xmm2\r |
| | 903 | movups 0(ctx), %xmm7\r |
| | 904 | aesdec %xmm6, %xmm1\r |
| | 905 | aesdec %xmm6, %xmm2\r |
| | 906 | aesdeclast %xmm7, %xmm1\r |
| | 907 | aesdeclast %xmm7, %xmm2\r |
| | 908 | movups 112(ctx), %xmm6\r |
| | 909 | movups 96(ctx), %xmm7\r |
| | 910 | #endif\r |
| | 911 | \r |
| | 912 | pxor iv, %xmm1 // obuf[0] ^= *iv; \r |
| | 913 | movups (ibuf), iv // ibuf[0]\r |
| | 914 | pxor iv, %xmm2 // obuf[1] ^= ibuf[0]\r |
| | 915 | movups 16(ibuf), iv // *iv = ibuf[1]\r |
| | 916 | \r |
| | 917 | movups %xmm1, (obuf) // write obuf[0]\r |
| | 918 | movups %xmm2, 16(obuf) // write obuf[1]\r |
| | 919 | \r |
| | 920 | add $32, ibuf // ibuf += 2\r |
| | 921 | add $32, obuf // obuf += 2\r |
| | 922 | \r |
| | 923 | 9:\r |
| | 924 | test $1, num_blk // check whether num_blk has residual 1 block\r |
| | 925 | je L_HW_cbc_done // if num_blk == 0, no need for residual processing code\r |
| | 926 | \r |
| | 927 | movups (ibuf), %xmm2 // tmp = ibuf\r |
| | 928 | // aes_decrypt\r |
| | 929 | pxor %xmm3, %xmm2\r |
| | 930 | aesdec %xmm4, %xmm2\r |
| | 931 | aesdec %xmm5, %xmm2\r |
| | 932 | aesdec %xmm6, %xmm2\r |
| | 933 | aesdec %xmm7, %xmm2\r |
| | 934 | #if defined __x86_64__\r |
| | 935 | aesdec %xmm8, %xmm2\r |
| | 936 | aesdec %xmm9, %xmm2\r |
| | 937 | aesdec %xmm10, %xmm2\r |
| | 938 | aesdec %xmm11, %xmm2\r |
| | 939 | aesdec %xmm12, %xmm2\r |
| | 940 | aesdeclast %xmm13, %xmm2\r |
| | 941 | #else\r |
| | 942 | movups 80(ctx), %xmm1\r |
| | 943 | aesdec %xmm1, %xmm2\r |
| | 944 | movups 64(ctx), %xmm1\r |
| | 945 | aesdec %xmm1, %xmm2\r |
| | 946 | movups 48(ctx), %xmm1\r |
| | 947 | aesdec %xmm1, %xmm2\r |
| | 948 | movups 32(ctx), %xmm1\r |
| | 949 | aesdec %xmm1, %xmm2\r |
| | 950 | movups 16(ctx), %xmm1\r |
| | 951 | aesdec %xmm1, %xmm2\r |
| | 952 | movups (ctx), %xmm1\r |
| | 953 | aesdeclast %xmm1, %xmm2\r |
| | 954 | #endif\r |
| | 955 | \r |
| | 956 | pxor iv, %xmm2 // *obuf ^= *iv; \r |
| | 957 | movups (ibuf), iv // *iv = *ibuf;\r |
| | 958 | movups %xmm2, (obuf) // write *obuf\r |
| | 959 | \r |
| | 960 | jmp L_HW_cbc_done\r |
| | 961 | \r |
| | 962 | //\r |
| | 963 | // aes-192 decrypt_cbc operation, after completion, branch to L_HW_cbc_done\r |
| | 964 | //\r |
| | 965 | \r |
| | 966 | L_decrypt_192:\r |
| | 967 | \r |
| | 968 | cmp $1, num_blk\r |
| | 969 | jl L_HW_cbc_done // if num_blk < 1, early return\r |
| | 970 | \r |
| | 971 | // aes-192 decryp expanded keys\r |
| | 972 | movups 192(ctx), %xmm3\r |
| | 973 | movups 176(ctx), %xmm4\r |
| | 974 | movups 160(ctx), %xmm5\r |
| | 975 | movups 144(ctx), %xmm6\r |
| | 976 | movups 128(ctx), %xmm7\r |
| | 977 | #if defined __x86_64__\r |
| | 978 | movups 112(ctx), %xmm8\r |
| | 979 | movups 96(ctx), %xmm9\r |
| | 980 | movups 80(ctx), %xmm10\r |
| | 981 | movups 64(ctx), %xmm11\r |
| | 982 | movups 48(ctx), %xmm12\r |
| | 983 | movups 32(ctx), %xmm13\r |
| | 984 | movups 16(ctx), %xmm14\r |
| | 985 | movups (ctx), %xmm15\r |
| | 986 | #endif\r |
| | 987 | \r |
| | 988 | // performs 4 block decryption in an iteration to exploit decrypt in parallel\r |
| | 989 | \r |
| | 990 | // while ((num_blk-=4)>=0) {\r |
| | 991 | // aes_decrypt(ibuf, obuf, ctx);\r |
| | 992 | // aes_decrypt(ibuf+1, obuf+1, ctx);\r |
| | 993 | // aes_decrypt(ibuf+2, obuf+2, ctx);\r |
| | 994 | // aes_decrypt(ibuf+3, obuf+3, ctx);\r |
| | 995 | // obuf[0] ^= *iv; obuf[1] ^= ibuf[1]; obuf[2] ^= ibuf[1]; obuf[3] ^= ibuf[2];\r |
| | 996 | // *iv = ibuf[3]; ibuf += 4; obuf += 4;\r |
| | 997 | // }\r |
| | 998 | \r |
| | 999 | sub $4, num_blk // pre decrement num_blk by 4\r |
| | 1000 | jl 9f // if num_blk < 4, skip the per-4-blocks processing code\r |
| | 1001 | 0:\r |
| | 1002 | \r |
| | 1003 | #if defined __x86_64__\r |
| | 1004 | \r |
| | 1005 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r |
| | 1006 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r |
| | 1007 | movups 32(ibuf), %xmm14 // tmp = 3rd ibuf\r |
| | 1008 | movups 48(ibuf), %xmm15 // tmp = 4th ibuf\r |
| | 1009 | \r |
| | 1010 | // aes_decrypt, for x86_64, the expanded keys are already stored in xmm3-xmm13\r |
| | 1011 | // use %xmm12/%xmm13 ts dynamic keys in the middle, restored afterwards\r |
| | 1012 | \r |
| | 1013 | // round 0 for 4 blocks\r |
| | 1014 | pxor %xmm3, %xmm1\r |
| | 1015 | pxor %xmm3, %xmm2\r |
| | 1016 | pxor %xmm3, %xmm14\r |
| | 1017 | pxor %xmm3, %xmm15\r |
| | 1018 | \r |
| | 1019 | // round 1 for 4 blocks\r |
| | 1020 | aesdec %xmm4, %xmm1\r |
| | 1021 | aesdec %xmm4, %xmm2\r |
| | 1022 | aesdec %xmm4, %xmm14\r |
| | 1023 | aesdec %xmm4, %xmm15\r |
| | 1024 | \r |
| | 1025 | // round 2 for 4 blocks\r |
| | 1026 | aesdec %xmm5, %xmm1\r |
| | 1027 | aesdec %xmm5, %xmm2\r |
| | 1028 | aesdec %xmm5, %xmm14\r |
| | 1029 | aesdec %xmm5, %xmm15\r |
| | 1030 | \r |
| | 1031 | // round 3 for 4 blocks\r |
| | 1032 | aesdec %xmm6, %xmm1\r |
| | 1033 | aesdec %xmm6, %xmm2\r |
| | 1034 | aesdec %xmm6, %xmm14\r |
| | 1035 | aesdec %xmm6, %xmm15\r |
| | 1036 | \r |
| | 1037 | // round 4 for 4 blocks\r |
| | 1038 | aesdec %xmm7, %xmm1\r |
| | 1039 | aesdec %xmm7, %xmm2\r |
| | 1040 | aesdec %xmm7, %xmm14\r |
| | 1041 | aesdec %xmm7, %xmm15\r |
| | 1042 | \r |
| | 1043 | // round 5 for 4 blocks\r |
| | 1044 | aesdec %xmm8, %xmm1\r |
| | 1045 | aesdec %xmm8, %xmm2\r |
| | 1046 | aesdec %xmm8, %xmm14\r |
| | 1047 | aesdec %xmm8, %xmm15\r |
| | 1048 | \r |
| | 1049 | // round 6 for 4 blocks\r |
| | 1050 | aesdec %xmm9, %xmm1\r |
| | 1051 | aesdec %xmm9, %xmm2\r |
| | 1052 | aesdec %xmm9, %xmm14\r |
| | 1053 | aesdec %xmm9, %xmm15\r |
| | 1054 | \r |
| | 1055 | // round 7 for 4 blocks\r |
| | 1056 | aesdec %xmm10, %xmm1\r |
| | 1057 | aesdec %xmm10, %xmm2\r |
| | 1058 | aesdec %xmm10, %xmm14\r |
| | 1059 | aesdec %xmm10, %xmm15\r |
| | 1060 | \r |
| | 1061 | // round 8 for 4 blocks\r |
| | 1062 | aesdec %xmm11, %xmm1\r |
| | 1063 | aesdec %xmm11, %xmm2\r |
| | 1064 | aesdec %xmm11, %xmm14\r |
| | 1065 | aesdec %xmm11, %xmm15\r |
| | 1066 | \r |
| | 1067 | // round 9 for 4 blocks\r |
| | 1068 | aesdec %xmm12, %xmm1\r |
| | 1069 | aesdec %xmm12, %xmm2\r |
| | 1070 | aesdec %xmm12, %xmm14\r |
| | 1071 | aesdec %xmm12, %xmm15\r |
| | 1072 | \r |
| | 1073 | movups 16(ctx), %xmm12\r |
| | 1074 | \r |
| | 1075 | // round A for 4 blocks\r |
| | 1076 | aesdec %xmm13, %xmm1\r |
| | 1077 | aesdec %xmm13, %xmm2\r |
| | 1078 | aesdec %xmm13, %xmm14\r |
| | 1079 | aesdec %xmm13, %xmm15\r |
| | 1080 | \r |
| | 1081 | movups (ctx), %xmm13\r |
| | 1082 | \r |
| | 1083 | // round B for 4 blocks\r |
| | 1084 | aesdec %xmm12, %xmm1\r |
| | 1085 | aesdec %xmm12, %xmm2\r |
| | 1086 | aesdec %xmm12, %xmm14\r |
| | 1087 | aesdec %xmm12, %xmm15\r |
| | 1088 | \r |
| | 1089 | movups 48(ctx), %xmm12 // restore %xmm12 to its original key\r |
| | 1090 | \r |
| | 1091 | // round C (last) for 4 blocks\r |
| | 1092 | aesdeclast %xmm13, %xmm1\r |
| | 1093 | aesdeclast %xmm13, %xmm2\r |
| | 1094 | aesdeclast %xmm13, %xmm14\r |
| | 1095 | aesdeclast %xmm13, %xmm15\r |
| | 1096 | \r |
| | 1097 | movups 32(ctx), %xmm13 // restore %xmm13 to its original key\r |
| | 1098 | \r |
| | 1099 | pxor iv, %xmm1 // obuf[0] ^= *iv; \r |
| | 1100 | movups (ibuf), iv // ibuf[0]\r |
| | 1101 | pxor iv, %xmm2 // obuf[1] ^= ibuf[0] \r |
| | 1102 | movups 16(ibuf), iv // ibuf[1]\r |
| | 1103 | pxor iv, %xmm14 // obuf[2] ^= ibuf[1] \r |
| | 1104 | movups 32(ibuf), iv // ibuf[2] \r |
| | 1105 | pxor iv, %xmm15 // obuf[3] ^= ibuf[2] \r |
| | 1106 | movups 48(ibuf), iv // *iv = ibuf[3] \r |
| | 1107 | \r |
| | 1108 | movups %xmm1, (obuf) // write 1st obuf\r |
| | 1109 | movups %xmm2, 16(obuf) // write 2nd obuf\r |
| | 1110 | movups %xmm14, 32(obuf) // write 3rd obuf\r |
| | 1111 | movups %xmm15, 48(obuf) // write 4th obuf\r |
| | 1112 | \r |
| | 1113 | add $64, ibuf // ibuf += 4; \r |
| | 1114 | add $64, obuf // obuf += 4; \r |
| | 1115 | \r |
| | 1116 | sub $4, num_blk // num_blk -= 4\r |
| | 1117 | jge 0b // if num_blk > 0, repeat the loop\r |
| | 1118 | \r |
| | 1119 | 9: add $4, num_blk // post incremtn num_blk by 4\r |
| | 1120 | je L_HW_cbc_done // if num_blk == 0, prepare to return \r |
| | 1121 | \r |
| | 1122 | movups 16(ctx), %xmm14 // restore %xmm14 to its key\r |
| | 1123 | movups (ctx), %xmm15 // restore %xmm15 to its key\r |
| | 1124 | \r |
| | 1125 | #else\r |
| | 1126 | \r |
| | 1127 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r |
| | 1128 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r |
| | 1129 | movups 32(ibuf), %xmm4 // tmp = 3rd ibuf\r |
| | 1130 | movups 48(ibuf), %xmm5 // tmp = 4th ibuf\r |
| | 1131 | \r |
| | 1132 | // aes_decrypt\r |
| | 1133 | // for i386, sequentially load expanded keys into xmm6/xmm7\r |
| | 1134 | movups 176(ctx), %xmm6\r |
| | 1135 | pxor %xmm3, %xmm1\r |
| | 1136 | pxor %xmm3, %xmm2\r |
| | 1137 | pxor %xmm3, %xmm4\r |
| | 1138 | pxor %xmm3, %xmm5\r |
| | 1139 | \r |
| | 1140 | movups 160(ctx), %xmm7\r |
| | 1141 | aesdec %xmm6, %xmm1\r |
| | 1142 | aesdec %xmm6, %xmm2\r |
| | 1143 | aesdec %xmm6, %xmm4\r |
| | 1144 | aesdec %xmm6, %xmm5\r |
| | 1145 | \r |
| | 1146 | movups 144(ctx), %xmm6\r |
| | 1147 | aesdec %xmm7, %xmm1\r |
| | 1148 | aesdec %xmm7, %xmm2\r |
| | 1149 | aesdec %xmm7, %xmm4\r |
| | 1150 | aesdec %xmm7, %xmm5\r |
| | 1151 | \r |
| | 1152 | movups 128(ctx), %xmm7\r |
| | 1153 | aesdec %xmm6, %xmm1\r |
| | 1154 | aesdec %xmm6, %xmm2\r |
| | 1155 | aesdec %xmm6, %xmm4\r |
| | 1156 | aesdec %xmm6, %xmm5\r |
| | 1157 | \r |
| | 1158 | movups 112(ctx), %xmm6\r |
| | 1159 | aesdec %xmm7, %xmm1\r |
| | 1160 | aesdec %xmm7, %xmm2\r |
| | 1161 | aesdec %xmm7, %xmm4\r |
| | 1162 | aesdec %xmm7, %xmm5\r |
| | 1163 | \r |
| | 1164 | movups 96(ctx), %xmm7\r |
| | 1165 | aesdec %xmm6, %xmm1\r |
| | 1166 | aesdec %xmm6, %xmm2\r |
| | 1167 | aesdec %xmm6, %xmm4\r |
| | 1168 | aesdec %xmm6, %xmm5\r |
| | 1169 | \r |
| | 1170 | movups 80(ctx), %xmm6\r |
| | 1171 | aesdec %xmm7, %xmm1\r |
| | 1172 | aesdec %xmm7, %xmm2\r |
| | 1173 | aesdec %xmm7, %xmm4\r |
| | 1174 | aesdec %xmm7, %xmm5\r |
| | 1175 | \r |
| | 1176 | movups 64(ctx), %xmm7\r |
| | 1177 | aesdec %xmm6, %xmm1\r |
| | 1178 | aesdec %xmm6, %xmm2\r |
| | 1179 | aesdec %xmm6, %xmm4\r |
| | 1180 | aesdec %xmm6, %xmm5\r |
| | 1181 | \r |
| | 1182 | movups 48(ctx), %xmm6\r |
| | 1183 | aesdec %xmm7, %xmm1\r |
| | 1184 | aesdec %xmm7, %xmm2\r |
| | 1185 | aesdec %xmm7, %xmm4\r |
| | 1186 | aesdec %xmm7, %xmm5\r |
| | 1187 | \r |
| | 1188 | movups 32(ctx), %xmm7\r |
| | 1189 | aesdec %xmm6, %xmm1\r |
| | 1190 | aesdec %xmm6, %xmm2\r |
| | 1191 | aesdec %xmm6, %xmm4\r |
| | 1192 | aesdec %xmm6, %xmm5\r |
| | 1193 | \r |
| | 1194 | movups 16(ctx), %xmm6\r |
| | 1195 | aesdec %xmm7, %xmm1\r |
| | 1196 | aesdec %xmm7, %xmm2\r |
| | 1197 | aesdec %xmm7, %xmm4\r |
| | 1198 | aesdec %xmm7, %xmm5\r |
| | 1199 | \r |
| | 1200 | movups 0(ctx), %xmm7\r |
| | 1201 | aesdec %xmm6, %xmm1\r |
| | 1202 | aesdec %xmm6, %xmm2\r |
| | 1203 | aesdec %xmm6, %xmm4\r |
| | 1204 | aesdec %xmm6, %xmm5\r |
| | 1205 | \r |
| | 1206 | aesdeclast %xmm7, %xmm1\r |
| | 1207 | aesdeclast %xmm7, %xmm2\r |
| | 1208 | aesdeclast %xmm7, %xmm4\r |
| | 1209 | aesdeclast %xmm7, %xmm5\r |
| | 1210 | \r |
| | 1211 | pxor iv, %xmm1 // 1st obuf ^= iv; \r |
| | 1212 | movups (ibuf), iv // 1st memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1213 | pxor iv, %xmm2 // 2nd obuf ^= iv; \r |
| | 1214 | movups 16(ibuf), iv // 2nd memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1215 | pxor iv, %xmm4 // 3rd obuf ^= iv; \r |
| | 1216 | movups 32(ibuf), iv // 3rd memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1217 | pxor iv, %xmm5 // 4th obuf ^= iv; \r |
| | 1218 | movups 48(ibuf), iv // 4th memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1219 | movups %xmm1, (obuf) // write 1st obuf\r |
| | 1220 | movups %xmm2, 16(obuf) // write 2nd obuf\r |
| | 1221 | movups %xmm4, 32(obuf) // write 3rd obuf\r |
| | 1222 | movups %xmm5, 48(obuf) // write 4th obuf\r |
| | 1223 | \r |
| | 1224 | add $64, ibuf // ibuf += AES_BLOCK_SIZE * 4; \r |
| | 1225 | add $64, obuf // obuf += AES_BLOCK_SIZE * 4; \r |
| | 1226 | \r |
| | 1227 | sub $4, num_blk // num_blk -= 4\r |
| | 1228 | jge 0b // if num_blk > 0, repeat the loop\r |
| | 1229 | \r |
| | 1230 | \r |
| | 1231 | 9: add $4, num_blk // post incremtn num_blk by 4\r |
| | 1232 | je L_HW_cbc_done // if num_blk == 0, no need for forthur processing code\r |
| | 1233 | \r |
| | 1234 | movups 176(ctx), %xmm4\r |
| | 1235 | movups 160(ctx), %xmm5\r |
| | 1236 | movups 144(ctx), %xmm6\r |
| | 1237 | movups 128(ctx), %xmm7\r |
| | 1238 | \r |
| | 1239 | #endif\r |
| | 1240 | \r |
| | 1241 | // per-block aes_decrypt_cbc loop\r |
| | 1242 | \r |
| | 1243 | 0:\r |
| | 1244 | movups (ibuf), %xmm2 // tmp = ibuf\r |
| | 1245 | \r |
| | 1246 | // aes_decrypt\r |
| | 1247 | pxor %xmm3, %xmm2\r |
| | 1248 | aesdec %xmm4, %xmm2\r |
| | 1249 | aesdec %xmm5, %xmm2\r |
| | 1250 | aesdec %xmm6, %xmm2\r |
| | 1251 | aesdec %xmm7, %xmm2\r |
| | 1252 | #if defined __x86_64__\r |
| | 1253 | aesdec %xmm8, %xmm2\r |
| | 1254 | aesdec %xmm9, %xmm2\r |
| | 1255 | aesdec %xmm10, %xmm2\r |
| | 1256 | aesdec %xmm11, %xmm2\r |
| | 1257 | aesdec %xmm12, %xmm2\r |
| | 1258 | aesdec %xmm13, %xmm2\r |
| | 1259 | aesdec %xmm14, %xmm2\r |
| | 1260 | aesdeclast %xmm15, %xmm2\r |
| | 1261 | #else\r |
| | 1262 | movups 112(ctx), %xmm1\r |
| | 1263 | aesdec %xmm1, %xmm2\r |
| | 1264 | movups 96(ctx), %xmm1\r |
| | 1265 | aesdec %xmm1, %xmm2\r |
| | 1266 | movups 80(ctx), %xmm1\r |
| | 1267 | aesdec %xmm1, %xmm2\r |
| | 1268 | movups 64(ctx), %xmm1\r |
| | 1269 | aesdec %xmm1, %xmm2\r |
| | 1270 | movups 48(ctx), %xmm1\r |
| | 1271 | aesdec %xmm1, %xmm2\r |
| | 1272 | movups 32(ctx), %xmm1\r |
| | 1273 | aesdec %xmm1, %xmm2\r |
| | 1274 | movups 16(ctx), %xmm1\r |
| | 1275 | aesdec %xmm1, %xmm2\r |
| | 1276 | movups (ctx), %xmm1\r |
| | 1277 | aesdeclast %xmm1, %xmm2\r |
| | 1278 | #endif\r |
| | 1279 | \r |
| | 1280 | pxor iv, %xmm2 // obuf ^= iv; \r |
| | 1281 | movups (ibuf), iv // memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1282 | \r |
| | 1283 | movups %xmm2, (obuf) // write obuf\r |
| | 1284 | \r |
| | 1285 | add $16, ibuf // ibuf += AES_BLOCK_SIZE; \r |
| | 1286 | add $16, obuf // obuf += AES_BLOCK_SIZE; \r |
| | 1287 | sub $1, num_blk // num_blk --\r |
| | 1288 | jg 0b // if num_blk > 0, repeat the loop\r |
| | 1289 | \r |
| | 1290 | jmp L_HW_cbc_done\r |
| | 1291 | \r |
| | 1292 | //\r |
| | 1293 | // aes-256 decrypt_cbc operation, after completion, branch to L_HW_cbc_done\r |
| | 1294 | //\r |
| | 1295 | \r |
| | 1296 | L_decrypt_256:\r |
| | 1297 | \r |
| | 1298 | cmp $1, num_blk\r |
| | 1299 | jl L_HW_cbc_done \r |
| | 1300 | \r |
| | 1301 | movups 224(ctx), %xmm3\r |
| | 1302 | movups 208(ctx), %xmm4\r |
| | 1303 | movups 192(ctx), %xmm5\r |
| | 1304 | movups 176(ctx), %xmm6\r |
| | 1305 | movups 160(ctx), %xmm7\r |
| | 1306 | #if defined __x86_64__\r |
| | 1307 | movups 144(ctx), %xmm8\r |
| | 1308 | movups 128(ctx), %xmm9\r |
| | 1309 | movups 112(ctx), %xmm10\r |
| | 1310 | movups 96(ctx), %xmm11\r |
| | 1311 | movups 80(ctx), %xmm12\r |
| | 1312 | movups 64(ctx), %xmm13\r |
| | 1313 | movups 48(ctx), %xmm14\r |
| | 1314 | movups 32(ctx), %xmm15\r |
| | 1315 | // movups 16(ctx), %xmm14\r |
| | 1316 | // movups (ctx), %xmm15\r |
| | 1317 | #endif\r |
| | 1318 | \r |
| | 1319 | #if defined __x86_64__\r |
| | 1320 | \r |
| | 1321 | sub $4, num_blk // pre decrement num_blk by 4\r |
| | 1322 | jl 9f // if num_blk < 4, skip the per-4-blocks processing code\r |
| | 1323 | 0:\r |
| | 1324 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r |
| | 1325 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r |
| | 1326 | movups 32(ibuf), %xmm14 // tmp = 3rd ibuf\r |
| | 1327 | movups 48(ibuf), %xmm15 // tmp = 4th ibuf\r |
| | 1328 | \r |
| | 1329 | // aes_decrypt, for x86_64, the expanded keys are already stored in xmm3-xmm13\r |
| | 1330 | pxor %xmm3, %xmm1\r |
| | 1331 | pxor %xmm3, %xmm2\r |
| | 1332 | pxor %xmm3, %xmm14\r |
| | 1333 | pxor %xmm3, %xmm15\r |
| | 1334 | \r |
| | 1335 | aesdec %xmm4, %xmm1\r |
| | 1336 | aesdec %xmm4, %xmm2\r |
| | 1337 | aesdec %xmm4, %xmm14\r |
| | 1338 | aesdec %xmm4, %xmm15\r |
| | 1339 | \r |
| | 1340 | aesdec %xmm5, %xmm1\r |
| | 1341 | aesdec %xmm5, %xmm2\r |
| | 1342 | aesdec %xmm5, %xmm14\r |
| | 1343 | aesdec %xmm5, %xmm15\r |
| | 1344 | \r |
| | 1345 | aesdec %xmm6, %xmm1\r |
| | 1346 | aesdec %xmm6, %xmm2\r |
| | 1347 | aesdec %xmm6, %xmm14\r |
| | 1348 | aesdec %xmm6, %xmm15\r |
| | 1349 | \r |
| | 1350 | aesdec %xmm7, %xmm1\r |
| | 1351 | aesdec %xmm7, %xmm2\r |
| | 1352 | aesdec %xmm7, %xmm14\r |
| | 1353 | aesdec %xmm7, %xmm15\r |
| | 1354 | \r |
| | 1355 | aesdec %xmm8, %xmm1\r |
| | 1356 | aesdec %xmm8, %xmm2\r |
| | 1357 | aesdec %xmm8, %xmm14\r |
| | 1358 | aesdec %xmm8, %xmm15\r |
| | 1359 | \r |
| | 1360 | aesdec %xmm9, %xmm1\r |
| | 1361 | aesdec %xmm9, %xmm2\r |
| | 1362 | aesdec %xmm9, %xmm14\r |
| | 1363 | aesdec %xmm9, %xmm15\r |
| | 1364 | \r |
| | 1365 | aesdec %xmm10, %xmm1\r |
| | 1366 | aesdec %xmm10, %xmm2\r |
| | 1367 | aesdec %xmm10, %xmm14\r |
| | 1368 | aesdec %xmm10, %xmm15\r |
| | 1369 | \r |
| | 1370 | aesdec %xmm11, %xmm1\r |
| | 1371 | aesdec %xmm11, %xmm2\r |
| | 1372 | aesdec %xmm11, %xmm14\r |
| | 1373 | aesdec %xmm11, %xmm15\r |
| | 1374 | \r |
| | 1375 | aesdec %xmm12, %xmm1\r |
| | 1376 | aesdec %xmm12, %xmm2\r |
| | 1377 | aesdec %xmm12, %xmm14\r |
| | 1378 | aesdec %xmm12, %xmm15\r |
| | 1379 | movups 48(ctx), %xmm12\r |
| | 1380 | \r |
| | 1381 | aesdec %xmm13, %xmm1\r |
| | 1382 | aesdec %xmm13, %xmm2\r |
| | 1383 | aesdec %xmm13, %xmm14\r |
| | 1384 | aesdec %xmm13, %xmm15\r |
| | 1385 | movups 32(ctx), %xmm13\r |
| | 1386 | \r |
| | 1387 | aesdec %xmm12, %xmm1\r |
| | 1388 | aesdec %xmm12, %xmm2\r |
| | 1389 | aesdec %xmm12, %xmm14\r |
| | 1390 | aesdec %xmm12, %xmm15\r |
| | 1391 | movups 16(ctx), %xmm12\r |
| | 1392 | \r |
| | 1393 | aesdec %xmm13, %xmm1\r |
| | 1394 | aesdec %xmm13, %xmm2\r |
| | 1395 | aesdec %xmm13, %xmm14\r |
| | 1396 | aesdec %xmm13, %xmm15\r |
| | 1397 | movups (ctx), %xmm13\r |
| | 1398 | \r |
| | 1399 | aesdec %xmm12, %xmm1\r |
| | 1400 | aesdec %xmm12, %xmm2\r |
| | 1401 | aesdec %xmm12, %xmm14\r |
| | 1402 | aesdec %xmm12, %xmm15\r |
| | 1403 | movups 80(ctx), %xmm12\r |
| | 1404 | \r |
| | 1405 | aesdeclast %xmm13, %xmm1\r |
| | 1406 | aesdeclast %xmm13, %xmm2\r |
| | 1407 | aesdeclast %xmm13, %xmm14\r |
| | 1408 | aesdeclast %xmm13, %xmm15\r |
| | 1409 | movups 64(ctx), %xmm13\r |
| | 1410 | \r |
| | 1411 | pxor iv, %xmm1 // obuf ^= iv; \r |
| | 1412 | movups (ibuf), iv // memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1413 | pxor iv, %xmm2 // obuf ^= iv; \r |
| | 1414 | movups 16(ibuf), iv // memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1415 | pxor iv, %xmm14 // obuf ^= iv; \r |
| | 1416 | movups 32(ibuf), iv // memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1417 | pxor iv, %xmm15 // obuf ^= iv; \r |
| | 1418 | movups 48(ibuf), iv // memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1419 | \r |
| | 1420 | movups %xmm1, (obuf) // write 1st obuf\r |
| | 1421 | movups %xmm2, 16(obuf) // write 2nd obuf\r |
| | 1422 | movups %xmm14, 32(obuf) // write 3rd obuf\r |
| | 1423 | movups %xmm15, 48(obuf) // write 4th obuf\r |
| | 1424 | \r |
| | 1425 | add $64, ibuf // ibuf += AES_BLOCK_SIZE*4; \r |
| | 1426 | add $64, obuf // obuf += AES_BLOCK_SIZE*4; \r |
| | 1427 | \r |
| | 1428 | sub $4, num_blk // num_blk -= 4\r |
| | 1429 | jge 0b // if num_blk > 0, repeat the loop\r |
| | 1430 | \r |
| | 1431 | 9: add $4, num_blk // post incremtn num_blk by 4\r |
| | 1432 | je L_HW_cbc_done // if num_blk == 0, no need for forthur processing code\r |
| | 1433 | \r |
| | 1434 | movups 48(ctx), %xmm14\r |
| | 1435 | movups 32(ctx), %xmm15\r |
| | 1436 | \r |
| | 1437 | #else\r |
| | 1438 | \r |
| | 1439 | sub $4, num_blk // pre decrement num_blk by 4\r |
| | 1440 | jl 9f // if num_blk < 4, skip the per-pair processing code\r |
| | 1441 | 0:\r |
| | 1442 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r |
| | 1443 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r |
| | 1444 | movups 32(ibuf), %xmm4 // tmp = 3rd ibuf\r |
| | 1445 | movups 48(ibuf), %xmm5 // tmp = 4th ibuf\r |
| | 1446 | \r |
| | 1447 | // aes_decrypt\r |
| | 1448 | // for i386, sequentially load expanded keys into xmm6/xmm7\r |
| | 1449 | movups 208(ctx), %xmm6\r |
| | 1450 | pxor %xmm3, %xmm1\r |
| | 1451 | pxor %xmm3, %xmm2\r |
| | 1452 | pxor %xmm3, %xmm4\r |
| | 1453 | pxor %xmm3, %xmm5\r |
| | 1454 | \r |
| | 1455 | movups 192(ctx), %xmm7\r |
| | 1456 | aesdec %xmm6, %xmm1\r |
| | 1457 | aesdec %xmm6, %xmm2\r |
| | 1458 | aesdec %xmm6, %xmm4\r |
| | 1459 | aesdec %xmm6, %xmm5\r |
| | 1460 | \r |
| | 1461 | movups 176(ctx), %xmm6\r |
| | 1462 | aesdec %xmm7, %xmm1\r |
| | 1463 | aesdec %xmm7, %xmm2\r |
| | 1464 | aesdec %xmm7, %xmm4\r |
| | 1465 | aesdec %xmm7, %xmm5\r |
| | 1466 | \r |
| | 1467 | movups 160(ctx), %xmm7\r |
| | 1468 | aesdec %xmm6, %xmm1\r |
| | 1469 | aesdec %xmm6, %xmm2\r |
| | 1470 | aesdec %xmm6, %xmm4\r |
| | 1471 | aesdec %xmm6, %xmm5\r |
| | 1472 | \r |
| | 1473 | movups 144(ctx), %xmm6\r |
| | 1474 | aesdec %xmm7, %xmm1\r |
| | 1475 | aesdec %xmm7, %xmm2\r |
| | 1476 | aesdec %xmm7, %xmm4\r |
| | 1477 | aesdec %xmm7, %xmm5\r |
| | 1478 | \r |
| | 1479 | movups 128(ctx), %xmm7\r |
| | 1480 | aesdec %xmm6, %xmm1\r |
| | 1481 | aesdec %xmm6, %xmm2\r |
| | 1482 | aesdec %xmm6, %xmm4\r |
| | 1483 | aesdec %xmm6, %xmm5\r |
| | 1484 | \r |
| | 1485 | movups 112(ctx), %xmm6\r |
| | 1486 | aesdec %xmm7, %xmm1\r |
| | 1487 | aesdec %xmm7, %xmm2\r |
| | 1488 | aesdec %xmm7, %xmm4\r |
| | 1489 | aesdec %xmm7, %xmm5\r |
| | 1490 | \r |
| | 1491 | movups 96(ctx), %xmm7\r |
| | 1492 | aesdec %xmm6, %xmm1\r |
| | 1493 | aesdec %xmm6, %xmm2\r |
| | 1494 | aesdec %xmm6, %xmm4\r |
| | 1495 | aesdec %xmm6, %xmm5\r |
| | 1496 | \r |
| | 1497 | movups 80(ctx), %xmm6\r |
| | 1498 | aesdec %xmm7, %xmm1\r |
| | 1499 | aesdec %xmm7, %xmm2\r |
| | 1500 | aesdec %xmm7, %xmm4\r |
| | 1501 | aesdec %xmm7, %xmm5\r |
| | 1502 | \r |
| | 1503 | movups 64(ctx), %xmm7\r |
| | 1504 | aesdec %xmm6, %xmm1\r |
| | 1505 | aesdec %xmm6, %xmm2\r |
| | 1506 | aesdec %xmm6, %xmm4\r |
| | 1507 | aesdec %xmm6, %xmm5\r |
| | 1508 | \r |
| | 1509 | movups 48(ctx), %xmm6\r |
| | 1510 | aesdec %xmm7, %xmm1\r |
| | 1511 | aesdec %xmm7, %xmm2\r |
| | 1512 | aesdec %xmm7, %xmm4\r |
| | 1513 | aesdec %xmm7, %xmm5\r |
| | 1514 | \r |
| | 1515 | movups 32(ctx), %xmm7\r |
| | 1516 | aesdec %xmm6, %xmm1\r |
| | 1517 | aesdec %xmm6, %xmm2\r |
| | 1518 | aesdec %xmm6, %xmm4\r |
| | 1519 | aesdec %xmm6, %xmm5\r |
| | 1520 | \r |
| | 1521 | movups 16(ctx), %xmm6\r |
| | 1522 | aesdec %xmm7, %xmm1\r |
| | 1523 | aesdec %xmm7, %xmm2\r |
| | 1524 | aesdec %xmm7, %xmm4\r |
| | 1525 | aesdec %xmm7, %xmm5\r |
| | 1526 | \r |
| | 1527 | movups 0(ctx), %xmm7\r |
| | 1528 | aesdec %xmm6, %xmm1\r |
| | 1529 | aesdec %xmm6, %xmm2\r |
| | 1530 | aesdec %xmm6, %xmm4\r |
| | 1531 | aesdec %xmm6, %xmm5\r |
| | 1532 | \r |
| | 1533 | aesdeclast %xmm7, %xmm1\r |
| | 1534 | aesdeclast %xmm7, %xmm2\r |
| | 1535 | aesdeclast %xmm7, %xmm4\r |
| | 1536 | aesdeclast %xmm7, %xmm5\r |
| | 1537 | \r |
| | 1538 | pxor iv, %xmm1 // 1st obuf ^= iv; \r |
| | 1539 | movups (ibuf), iv // 1st memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1540 | pxor iv, %xmm2 // 2nd obuf ^= iv; \r |
| | 1541 | movups 16(ibuf), iv // 2nd memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1542 | pxor iv, %xmm4 // 3rd obuf ^= iv; \r |
| | 1543 | movups 32(ibuf), iv // 3rd memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1544 | pxor iv, %xmm5 // 4th obuf ^= iv; \r |
| | 1545 | movups 48(ibuf), iv // 4th memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1546 | movups %xmm1, (obuf) // write 1st obuf\r |
| | 1547 | movups %xmm2, 16(obuf) // write 2nd obuf\r |
| | 1548 | movups %xmm4, 32(obuf) // write 3rd obuf\r |
| | 1549 | movups %xmm5, 48(obuf) // write 4th obuf\r |
| | 1550 | \r |
| | 1551 | add $64, ibuf // ibuf += AES_BLOCK_SIZE * 4; \r |
| | 1552 | add $64, obuf // obuf += AES_BLOCK_SIZE * 4; \r |
| | 1553 | \r |
| | 1554 | sub $4, num_blk // num_blk -= 4\r |
| | 1555 | jge 0b // if num_blk > 0, repeat the loop\r |
| | 1556 | \r |
| | 1557 | \r |
| | 1558 | 9: add $4, num_blk // post incremtn num_blk by 4\r |
| | 1559 | je L_HW_cbc_done // if num_blk == 0, no need for forthur processing code\r |
| | 1560 | \r |
| | 1561 | movups 208(ctx), %xmm4\r |
| | 1562 | movups 192(ctx), %xmm5\r |
| | 1563 | movups 176(ctx), %xmm6\r |
| | 1564 | movups 160(ctx), %xmm7\r |
| | 1565 | \r |
| | 1566 | #endif\r |
| | 1567 | \r |
| | 1568 | 0:\r |
| | 1569 | movups (ibuf), %xmm2 // tmp = ibuf\r |
| | 1570 | \r |
| | 1571 | // aes_decrypt\r |
| | 1572 | pxor %xmm3, %xmm2\r |
| | 1573 | aesdec %xmm4, %xmm2\r |
| | 1574 | aesdec %xmm5, %xmm2\r |
| | 1575 | aesdec %xmm6, %xmm2\r |
| | 1576 | aesdec %xmm7, %xmm2\r |
| | 1577 | #if defined __x86_64__\r |
| | 1578 | aesdec %xmm8, %xmm2\r |
| | 1579 | aesdec %xmm9, %xmm2\r |
| | 1580 | aesdec %xmm10, %xmm2\r |
| | 1581 | aesdec %xmm11, %xmm2\r |
| | 1582 | aesdec %xmm12, %xmm2\r |
| | 1583 | aesdec %xmm13, %xmm2\r |
| | 1584 | aesdec %xmm14, %xmm2\r |
| | 1585 | aesdec %xmm15, %xmm2\r |
| | 1586 | #else\r |
| | 1587 | movups 144(ctx), %xmm1\r |
| | 1588 | aesdec %xmm1, %xmm2\r |
| | 1589 | movups 128(ctx), %xmm1\r |
| | 1590 | aesdec %xmm1, %xmm2\r |
| | 1591 | movups 112(ctx), %xmm1\r |
| | 1592 | aesdec %xmm1, %xmm2\r |
| | 1593 | movups 96(ctx), %xmm1\r |
| | 1594 | aesdec %xmm1, %xmm2\r |
| | 1595 | movups 80(ctx), %xmm1\r |
| | 1596 | aesdec %xmm1, %xmm2\r |
| | 1597 | movups 64(ctx), %xmm1\r |
| | 1598 | aesdec %xmm1, %xmm2\r |
| | 1599 | movups 48(ctx), %xmm1\r |
| | 1600 | aesdec %xmm1, %xmm2\r |
| | 1601 | movups 32(ctx), %xmm1\r |
| | 1602 | aesdec %xmm1, %xmm2\r |
| | 1603 | #endif\r |
| | 1604 | movups 16(ctx), %xmm1\r |
| | 1605 | aesdec %xmm1, %xmm2\r |
| | 1606 | movups (ctx), %xmm1\r |
| | 1607 | aesdeclast %xmm1, %xmm2\r |
| | 1608 | \r |
| | 1609 | pxor iv, %xmm2 // obuf ^= iv; \r |
| | 1610 | movups (ibuf), iv // memcpy(iv, tmp, AES_BLOCK_SIZE);\r |
| | 1611 | \r |
| | 1612 | movups %xmm2, (obuf) // write obuf\r |
| | 1613 | \r |
| | 1614 | add $16, ibuf // ibuf += AES_BLOCK_SIZE; \r |
| | 1615 | add $16, obuf // obuf += AES_BLOCK_SIZE; \r |
| | 1616 | sub $1, num_blk // num_blk --\r |
| | 1617 | jg 0b // if num_blk > 0, repeat the loop\r |
| | 1618 | \r |
| | 1619 | jmp L_HW_cbc_done\r |
| | 1620 | \r |
| | 1621 | //\r |
| | 1622 | // --------- END of aes_decrypt_cbc_hw -------------------\r |
| | 1623 | //\r |
projects / apple / xnu.git / blame_incremental