projects / apple / xnu.git / blame
| Commit | Line | Data |
|---|---|---|
| 6d2010ae A |
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 | /* modified 3/5/10 cclee */\r | |
| 41 | /* Clean up those related to VIA ACE and hand optimize aes_cbc_encrypt and aes_cbc_decrypt */\r | |
| 42 | /* move the xmm registers save/restore originally inside the callee functions into these 2 caller functions */\r | |
| 43 | \r | |
| 44 | /* HW-AES specific implementation cclee 3-12-10 */\r | |
| 45 | /* In aes_encrypt_cbc and aes_decrypt_cbc, __cpu_capabilities is polled, \r | |
| 46 | and if kHasAES is detected, branch to the hw-specific functions here */\r | |
| 47 | \r | |
| 48 | \r | |
| 49 | /* \r | |
| 50 | This files defines _aes_encrypt_cbc_hw and _aes_decrypt_cbc_hw --- Intel Westmere HW AES-based implementation\r | |
| 51 | of _aes_encrypt_cbc and _aes_decrypt_cbc. \r | |
| 52 | \r | |
| 53 | These 2 functions SHOULD BE entried ONLY after the AES HW is verified to be available. \r | |
| 54 | They SHOULD NOT be called without AES HW detection. It might cause xnu to crash.\r | |
| 55 | \r | |
| 56 | The AES HW is detected 1st thing in \r | |
| 57 | _aes_encrypt_cbc (aes_modes_asm.s) \r | |
| 58 | _aes_decrypt_cbc (aes_modes_asm.s)\r | |
| 59 | and, if AES HW is detected, branch without link (ie, jump) to the functions here.\r | |
| 60 | \r | |
| 61 | The implementation here follows the examples in an Intel White Paper\r | |
| 62 | "Intel Advanced Encryption Standard (AES) Instruction Set" Rev.2 01\r | |
| 63 | \r | |
| 64 | Note: Rev. 03 Final 2010 01 26 is available. Looks like some code change from Rev.2 01\r | |
| 65 | \r | |
| 66 | cclee 3-13-10\r | |
| 67 | */\r | |
| 68 | \r | |
| 69 | /* \r | |
| 70 | The function _aes_decrypt_cbc_hw previously simply serially decrypts block by block\r | |
| 71 | in our group meeting, Eric/Ali suggested that I perhaps should take a look of combining multiple blocks\r | |
| 72 | in a loop and interleaving multiple aesdec instructions to absorb/hide stalls to improve the decrypt thoughput.\r | |
| 73 | \r | |
| 74 | The idea was actually described in the Intel AES Instruction Set White Paper (Rev. 2.0 page 53-55) \r | |
| 75 | \r | |
| 76 | This modification interleaves the aesdec/aesdeclast instructions for 4 blocks in cbc mode.\r | |
| 77 | On a K18 (2.4GHz core-i5/2.66GHz core-i7), the x86_64 decrypt throughput (in xnu-iokit) has been improved\r | |
| 78 | from 1180/1332 to 1667/1858 MBytes/sec. This is approximately 1.40 times speedup in the decryption.\r | |
| 79 | The encrypt throughput is not changed. \r | |
| 80 | \r | |
| 81 | I also enhanced the assembly code comments.\r | |
| 82 | \r | |
| 83 | cclee-4-30-10 (Do you know 4-30 is National Honesty Day in the US? No need to know. I've been honest all the time.)\r | |
| 84 | \r | |
| 85 | */\r | |
| 86 | \r | |
| 87 | /* ---------------------------------------------------------------------------------------------------------------- \r | |
| 88 | \r | |
| 89 | aes_encrypt_cbc function (see aes_modes.c or aes_modes_asm.s) :\r | |
| 90 | \r | |
| 91 | For simplicity, I am assuming all variables are in 128-bit data type.\r | |
| 92 | \r | |
| 93 | aes_rval aes_encrypt_cbc(const __m128 *ibuf, __m128 *iv, int num_blk, __m128 *obuf, const aes_encrypt_ctx *ctx)\r | |
| 94 | {\r | |
| 95 | while(num_blk--) {\r | |
| 96 | *iv ^= *ibuf++;\r | |
| 97 | aes_encrypt(iv, iv, ctx);\r | |
| 98 | *obuf++ = *iv;\r | |
| 99 | }\r | |
| 100 | return 0;\r | |
| 101 | }\r | |
| 102 | \r | |
| 103 | The following is an implementation of this function using Intel AESNI.\r | |
| 104 | This function _aes_encrypt_cbc_hw SHOULD NOT be called directly. \r | |
| 105 | Developer should still call _aes_encrypt_cbc (in aes_modes_asm.s) which will poll cpu_capabilities and branch\r | |
| 106 | to this aesni-based function should it detecs that aesni is available.\r | |
| 107 | Blindly call this function SURELY will cause a CRASH on systems with no aesni support. \r | |
| 108 | \r | |
| 109 | Note that each block starts with *iv, which is the output of the previous block. Therefore, the cbc blocks\r | |
| 110 | are serially chained. This prevents us from arranging several blocks for encryption in parallel.\r | |
| 111 | \r | |
| 112 | ----------------------------------------------------------------------------------------------------------------*/\r | |
| 113 | \r | |
| 114 | .text\r | |
| 115 | .align 4,0x90\r | |
| 116 | .globl _aes_encrypt_cbc_hw\r | |
| 117 | _aes_encrypt_cbc_hw:\r | |
| 118 | \r | |
| 119 | // push/save registers for local use\r | |
| 120 | #if defined __i386__\r | |
| 121 | \r | |
| 122 | push %ebp\r | |
| 123 | movl %esp, %ebp\r | |
| 124 | push %ebx\r | |
| 125 | push %edi\r | |
| 126 | \r | |
| 127 | #define sp %esp\r | |
| 128 | \r | |
| 129 | #else // __x86_64__\r | |
| 130 | \r | |
| 131 | push %rbp\r | |
| 132 | mov %rsp, %rbp\r | |
| 133 | push %rbx\r | |
| 134 | push %r13\r | |
| 135 | push %r14\r | |
| 136 | push %r15\r | |
| 137 | \r | |
| 138 | #define sp %rsp\r | |
| 139 | \r | |
| 140 | #endif\r | |
| 141 | \r | |
| 142 | // if this is kernel code, need to save used xmm registers\r | |
| 143 | #ifdef KERNEL\r | |
| 144 | \r | |
| 145 | #if defined __i386__\r | |
| 146 | sub $(8*16), %esp // for possible xmm0-xmm7 save/restore\r | |
| 147 | #else\r | |
| 148 | sub $(16*16), %rsp // xmm0-xmm15 save/restore \r | |
| 149 | #endif\r | |
| 150 | \r | |
| 151 | movaps %xmm0, (sp)\r | |
| 152 | movaps %xmm1, 16(sp)\r | |
| 153 | movaps %xmm2, 32(sp)\r | |
| 154 | movaps %xmm3, 48(sp)\r | |
| 155 | movaps %xmm4, 64(sp)\r | |
| 156 | movaps %xmm5, 80(sp)\r | |
| 157 | movaps %xmm6, 96(sp)\r | |
| 158 | movaps %xmm7, 112(sp)\r | |
| 159 | #if defined __x86_64__\r | |
| 160 | movaps %xmm8, 16*8(sp)\r | |
| 161 | movaps %xmm9, 16*9(sp)\r | |
| 162 | movaps %xmm10, 16*10(sp)\r | |
| 163 | movaps %xmm11, 16*11(sp)\r | |
| 164 | movaps %xmm12, 16*12(sp)\r | |
| 165 | movaps %xmm13, 16*13(sp)\r | |
| 166 | movaps %xmm14, 16*14(sp)\r | |
| 167 | movaps %xmm15, 16*15(sp)\r | |
| 168 | #endif // __x86_64__\r | |
| 169 | \r | |
| 170 | #endif // KERNEL\r | |
| 171 | \r | |
| 172 | #define iv %xmm0\r | |
| 173 | \r | |
| 174 | #ifdef __i386__\r | |
| 175 | \r | |
| 176 | mov 12(%ebp), %eax // in_iv\r | |
| 177 | mov 24(%ebp), %edx // ctx\r | |
| 178 | movups (%eax), iv // iv = in_iv \r | |
| 179 | mov 8(%ebp), %ebx // ibuf\r | |
| 180 | mov 16(%ebp), %ecx // num_blk\r | |
| 181 | mov 20(%ebp), %edi // obuf\r | |
| 182 | \r | |
| 183 | #define ibuf %ebx\r | |
| 184 | #define obuf %edi\r | |
| 185 | #define num_blk %ecx \r | |
| 186 | #define ctx %edx\r | |
| 187 | \r | |
| 188 | #else\r | |
| 189 | \r | |
| 190 | mov %rdi, %rbx // ibuf\r | |
| 191 | movups (%rsi), iv // iv = in_iv\r | |
| 192 | mov %rdx, %r13 // num_blk\r | |
| 193 | mov %rcx, %r14 // obuf\r | |
| 194 | mov %r8, %r15 // ctx \r | |
| 195 | \r | |
| 196 | #define ibuf %rbx\r | |
| 197 | #define num_blk %r13d\r | |
| 198 | #define obuf %r14 \r | |
| 199 | #define ctx %r15\r | |
| 200 | \r | |
| 201 | #endif\r | |
| 202 | \r | |
| 203 | mov 240(ctx), %eax // aes length\r | |
| 204 | cmp $160, %eax // aes-128 encrypt ?\r | |
| 205 | je L_encrypt_128\r | |
| 206 | cmp $192, %eax // aes-192 encrypt ?\r | |
| 207 | je L_encrypt_192\r | |
| 208 | cmp $224, %eax // aes-256 encrypt ?\r | |
| 209 | je L_encrypt_256\r | |
| 210 | mov $-1, %eax // return error\r | |
| 211 | jmp L_error \r | |
| 212 | \r | |
| 213 | //\r | |
| 214 | // aes-128 encrypt_cbc operation, up to L_HW_cbc_done\r | |
| 215 | //\r | |
| 216 | \r | |
| 217 | L_encrypt_128:\r | |
| 218 | \r | |
| 219 | cmp $1, num_blk // check number of block\r | |
| 220 | jl L_HW_cbc_done // should it be less than 1, nothing to do\r | |
| 221 | \r | |
| 222 | movups (ctx), %xmm2 // key0\r | |
| 223 | movups 16(ctx), %xmm3 // key1\r | |
| 224 | movups 32(ctx), %xmm4 // key2\r | |
| 225 | movups 48(ctx), %xmm5 // key3\r | |
| 226 | movups 64(ctx), %xmm6 // key4\r | |
| 227 | movups 80(ctx), %xmm7 // key5\r | |
| 228 | #if defined __x86_64__\r | |
| 229 | movups 96(ctx), %xmm8 // key6\r | |
| 230 | movups 112(ctx), %xmm9 // key7\r | |
| 231 | movups 128(ctx), %xmm10 // key8\r | |
| 232 | movups 144(ctx), %xmm11 // key9\r | |
| 233 | movups 160(ctx), %xmm12 // keyA\r | |
| 234 | #endif\r | |
| 235 | \r | |
| 236 | // while (num_blk--) {\r | |
| 237 | // *iv ^= *ibuf++;\r | |
| 238 | // aes_encrypt(iv, iv, ctx);\r | |
| 239 | // *obuf++ = *iv;\r | |
| 240 | // }\r | |
| 241 | 0:\r | |
| 242 | movups (ibuf), %xmm1 // *ibuf\r | |
| 243 | pxor %xmm2, iv // 1st instruction inside aes_encrypt\r | |
| 244 | pxor %xmm1, iv // *iv ^= *ibuf\r | |
| 245 | \r | |
| 246 | // finishing up the rest of aes_encrypt\r | |
| 247 | aesenc %xmm3, iv\r | |
| 248 | aesenc %xmm4, iv\r | |
| 249 | aesenc %xmm5, iv\r | |
| 250 | aesenc %xmm6, iv\r | |
| 251 | aesenc %xmm7, iv\r | |
| 252 | #if defined __x86_64__\r | |
| 253 | aesenc %xmm8, iv\r | |
| 254 | aesenc %xmm9, iv\r | |
| 255 | aesenc %xmm10, iv\r | |
| 256 | aesenc %xmm11, iv\r | |
| 257 | aesenclast %xmm12, iv\r | |
| 258 | #else\r | |
| 259 | movups 96(ctx), %xmm1 // key6\r | |
| 260 | aesenc %xmm1, iv\r | |
| 261 | movups 112(ctx), %xmm1 // key7\r | |
| 262 | aesenc %xmm1, iv\r | |
| 263 | movups 128(ctx), %xmm1 // key8\r | |
| 264 | aesenc %xmm1, iv\r | |
| 265 | movups 144(ctx), %xmm1 // key9\r | |
| 266 | aesenc %xmm1, iv\r | |
| 267 | movups 160(ctx), %xmm1 // keyA\r | |
| 268 | aesenclast %xmm1, iv\r | |
| 269 | #endif\r | |
| 270 | \r | |
| 271 | movups iv, (obuf) // *obuf = *iv;\r | |
| 272 | add $16, obuf // obuf++;\r | |
| 273 | add $16, ibuf // ibuf++;\r | |
| 274 | sub $1, num_blk // num_blk --\r | |
| 275 | jg 0b // if num_blk > 0, repeat the loop\r | |
| 276 | \r | |
| 277 | // the following will be branched to from all other cases (encrypt/decrypt 128/192/256)\r | |
| 278 | \r | |
| 279 | L_HW_cbc_done:\r | |
| 280 | \r | |
| 281 | xor %eax, %eax // to return CRYPT_OK\r | |
| 282 | \r | |
| 283 | L_error:\r | |
| 284 | \r | |
| 285 | // if kernel, restore xmm registers\r | |
| 286 | #ifdef KERNEL \r | |
| 287 | movaps 0(sp), %xmm0\r | |
| 288 | movaps 16(sp), %xmm1\r | |
| 289 | movaps 32(sp), %xmm2\r | |
| 290 | movaps 48(sp), %xmm3\r | |
| 291 | movaps 64(sp), %xmm4\r | |
| 292 | movaps 80(sp), %xmm5\r | |
| 293 | movaps 96(sp), %xmm6\r | |
| 294 | movaps 112(sp), %xmm7\r | |
| 295 | #if defined __x86_64__\r | |
| 296 | movaps 16*8(sp), %xmm8\r | |
| 297 | movaps 16*9(sp), %xmm9\r | |
| 298 | movaps 16*10(sp), %xmm10\r | |
| 299 | movaps 16*11(sp), %xmm11\r | |
| 300 | movaps 16*12(sp), %xmm12\r | |
| 301 | movaps 16*13(sp), %xmm13\r | |
| 302 | movaps 16*14(sp), %xmm14\r | |
| 303 | movaps 16*15(sp), %xmm15\r | |
| 304 | #endif // __x86_64__\r | |
| 305 | #endif // KERNEL\r | |
| 306 | \r | |
| 307 | // release used stack memory, restore used callee-saved registers, and return \r | |
| 308 | #if defined __i386__\r | |
| 309 | #ifdef KERNEL\r | |
| 310 | add $(8*16), %esp\r | |
| 311 | #endif\r | |
| 312 | pop %edi\r | |
| 313 | pop %ebx\r | |
| 314 | #else\r | |
| 315 | #ifdef KERNEL\r | |
| 316 | add $(16*16), %rsp \r | |
| 317 | #endif\r | |
| 318 | pop %r15\r | |
| 319 | pop %r14\r | |
| 320 | pop %r13\r | |
| 321 | pop %rbx\r | |
| 322 | #endif\r | |
| 323 | leave\r | |
| 324 | ret\r | |
| 325 | \r | |
| 326 | //\r | |
| 327 | // aes-192 encrypt_cbc operation, after completion, branch to L_HW_cbc_done\r | |
| 328 | //\r | |
| 329 | \r | |
| 330 | L_encrypt_192:\r | |
| 331 | \r | |
| 332 | cmp $1, num_blk // check number of block\r | |
| 333 | jl L_HW_cbc_done // should it be less than 1, nothing to do\r | |
| 334 | \r | |
| 335 | movups (ctx), %xmm2 // key0\r | |
| 336 | movups 16(ctx), %xmm3 // key1\r | |
| 337 | movups 32(ctx), %xmm4 // key2\r | |
| 338 | movups 48(ctx), %xmm5 // key3\r | |
| 339 | movups 64(ctx), %xmm6 // key4\r | |
| 340 | movups 80(ctx), %xmm7 // key5\r | |
| 341 | #if defined __x86_64__\r | |
| 342 | movups 96(ctx), %xmm8 // key6\r | |
| 343 | movups 112(ctx), %xmm9 // key7\r | |
| 344 | movups 128(ctx), %xmm10 // key8\r | |
| 345 | movups 144(ctx), %xmm11 // key9\r | |
| 346 | movups 160(ctx), %xmm12 // keyA\r | |
| 347 | movups 176(ctx), %xmm13 // keyB\r | |
| 348 | movups 192(ctx), %xmm14 // keyC\r | |
| 349 | #endif\r | |
| 350 | \r | |
| 351 | // while (num_blk--) {\r | |
| 352 | // *iv ^= *ibuf++;\r | |
| 353 | // aes_encrypt(iv, iv, ctx);\r | |
| 354 | // *obuf++ = *iv;\r | |
| 355 | // }\r | |
| 356 | 0:\r | |
| 357 | movups (ibuf), %xmm1 // *ibuf\r | |
| 358 | pxor %xmm1, iv // *iv ^= ibuf\r | |
| 359 | \r | |
| 360 | // aes_encrypt(iv, iv, ctx);\r | |
| 361 | \r | |
| 362 | pxor %xmm2, iv\r | |
| 363 | aesenc %xmm3, iv\r | |
| 364 | aesenc %xmm4, iv\r | |
| 365 | aesenc %xmm5, iv\r | |
| 366 | aesenc %xmm6, iv\r | |
| 367 | aesenc %xmm7, iv\r | |
| 368 | #if defined __x86_64__\r | |
| 369 | aesenc %xmm8, iv\r | |
| 370 | aesenc %xmm9, iv\r | |
| 371 | aesenc %xmm10, iv\r | |
| 372 | aesenc %xmm11, iv\r | |
| 373 | aesenc %xmm12, iv\r | |
| 374 | aesenc %xmm13, iv\r | |
| 375 | aesenclast %xmm14, iv\r | |
| 376 | #else\r | |
| 377 | movups 96(ctx), %xmm1\r | |
| 378 | aesenc %xmm1, iv\r | |
| 379 | movups 112(ctx), %xmm1\r | |
| 380 | aesenc %xmm1, iv\r | |
| 381 | movups 128(ctx), %xmm1\r | |
| 382 | aesenc %xmm1, iv\r | |
| 383 | movups 144(ctx), %xmm1\r | |
| 384 | aesenc %xmm1, iv\r | |
| 385 | movups 160(ctx), %xmm1\r | |
| 386 | aesenc %xmm1, iv\r | |
| 387 | movups 176(ctx), %xmm1\r | |
| 388 | aesenc %xmm1, iv\r | |
| 389 | movups 192(ctx), %xmm1\r | |
| 390 | aesenclast %xmm1, iv\r | |
| 391 | #endif\r | |
| 392 | \r | |
| 393 | movups iv, (obuf) // *obuf = *iv;\r | |
| 394 | add $16, ibuf // ibuf++\r | |
| 395 | add $16, obuf // obuf++\r | |
| 396 | \r | |
| 397 | sub $1, num_blk // num_blk --\r | |
| 398 | jg 0b // if num_blk > 0, repeat the loop\r | |
| 399 | \r | |
| 400 | jmp L_HW_cbc_done // share with the common exit code\r | |
| 401 | \r | |
| 402 | //\r | |
| 403 | // aes-256 encrypt_cbc operation, after completion, branch to L_HW_cbc_done\r | |
| 404 | //\r | |
| 405 | \r | |
| 406 | L_encrypt_256:\r | |
| 407 | \r | |
| 408 | cmp $1, num_blk // check number of block\r | |
| 409 | jl L_HW_cbc_done // should it be less than 1, nothing to do\r | |
| 410 | \r | |
| 411 | movups (ctx), %xmm2 // key0\r | |
| 412 | movups 16(ctx), %xmm3 // key1\r | |
| 413 | movups 32(ctx), %xmm4 // key2\r | |
| 414 | movups 48(ctx), %xmm5 // key3\r | |
| 415 | movups 64(ctx), %xmm6 // key4\r | |
| 416 | movups 80(ctx), %xmm7 // key5\r | |
| 417 | #if defined __x86_64__\r | |
| 418 | movups 96(ctx), %xmm8 // key6\r | |
| 419 | movups 112(ctx), %xmm9 // key7\r | |
| 420 | movups 128(ctx), %xmm10 // key8\r | |
| 421 | movups 144(ctx), %xmm11 // key9\r | |
| 422 | movups 160(ctx), %xmm12 // keyA\r | |
| 423 | movups 176(ctx), %xmm13 // keyB\r | |
| 424 | movups 192(ctx), %xmm14 // keyC\r | |
| 425 | movups 208(ctx), %xmm15 // keyD\r | |
| 426 | // movups 224(ctx), %xmm1 // keyE\r | |
| 427 | #endif\r | |
| 428 | \r | |
| 429 | // while (num_blk--) {\r | |
| 430 | // *iv ^= *ibuf++;\r | |
| 431 | // aes_encrypt(iv, iv, ctx);\r | |
| 432 | // *obuf++ = *iv;\r | |
| 433 | // }\r | |
| 434 | 0:\r | |
| 435 | movups (ibuf), %xmm1 // *ibuf\r | |
| 436 | pxor %xmm1, iv // *iv ^= ibuf\r | |
| 437 | \r | |
| 438 | // aes_encrypt(iv, iv, ctx);\r | |
| 439 | pxor %xmm2, iv\r | |
| 440 | aesenc %xmm3, iv\r | |
| 441 | aesenc %xmm4, iv\r | |
| 442 | aesenc %xmm5, iv\r | |
| 443 | aesenc %xmm6, iv\r | |
| 444 | aesenc %xmm7, iv\r | |
| 445 | #if defined __x86_64__\r | |
| 446 | movups 224(ctx), %xmm1 // keyE\r | |
| 447 | aesenc %xmm8, iv\r | |
| 448 | aesenc %xmm9, iv\r | |
| 449 | aesenc %xmm10, iv\r | |
| 450 | aesenc %xmm11, iv\r | |
| 451 | aesenc %xmm12, iv\r | |
| 452 | aesenc %xmm13, iv\r | |
| 453 | aesenc %xmm14, iv\r | |
| 454 | aesenc %xmm15, iv\r | |
| 455 | aesenclast %xmm1, iv\r | |
| 456 | #else\r | |
| 457 | movups 96(ctx), %xmm1 // key6\r | |
| 458 | aesenc %xmm1, iv\r | |
| 459 | movups 112(ctx), %xmm1 // key7\r | |
| 460 | aesenc %xmm1, iv\r | |
| 461 | movups 128(ctx), %xmm1 // key8\r | |
| 462 | aesenc %xmm1, iv\r | |
| 463 | movups 144(ctx), %xmm1 // key9\r | |
| 464 | aesenc %xmm1, iv\r | |
| 465 | movups 160(ctx), %xmm1 // keyA\r | |
| 466 | aesenc %xmm1, iv\r | |
| 467 | movups 176(ctx), %xmm1 // keyB\r | |
| 468 | aesenc %xmm1, iv\r | |
| 469 | movups 192(ctx), %xmm1 // keyC\r | |
| 470 | aesenc %xmm1, iv\r | |
| 471 | movups 208(ctx), %xmm1 // keyD\r | |
| 472 | aesenc %xmm1, iv\r | |
| 473 | movups 224(ctx), %xmm1 // keyE\r | |
| 474 | aesenclast %xmm1, iv\r | |
| 475 | #endif\r | |
| 476 | \r | |
| 477 | movups iv, (obuf) // *obuf = *iv;\r | |
| 478 | add $16, ibuf // ibuf++\r | |
| 479 | add $16, obuf // obuf++\r | |
| 480 | \r | |
| 481 | sub $1, num_blk // num_blk --\r | |
| 482 | jg 0b // if num_blk > 0, repeat the loop\r | |
| 483 | \r | |
| 484 | jmp L_HW_cbc_done // share with the common exit code\r | |
| 485 | \r | |
| 486 | \r | |
| 487 | \r | |
| 488 | //\r | |
| 489 | // --------- END of aes_encrypt_cbc_hw -------------------\r | |
| 490 | //\r | |
| 491 | \r | |
| 492 | \r | |
| 493 | /* ---------------------------------------------------------------------------------------------------------------- \r | |
| 494 | \r | |
| 495 | aes_decrypt_cbc function (see aes_modes.c or aes_modes_asm.s) :\r | |
| 496 | \r | |
| 497 | For simplicity, I am assuming all variables are in 128-bit data type.\r | |
| 498 | \r | |
| 499 | aes_rval aes_decrypt_cbc(const __m128 *ibuf, __m128 *iv, int num_blk, __m128 *obuf, const aes_decrypt_ctx *ctx)\r | |
| 500 | {\r | |
| 501 | while(num_blk--) {\r | |
| 502 | aes_decrypt(ibuf, obuf, ctx);\r | |
| 503 | *obuf++ ^= *iv;\r | |
| 504 | *iv = *ibuf++;\r | |
| 505 | }\r | |
| 506 | return 0;\r | |
| 507 | }\r | |
| 508 | \r | |
| 509 | The following is an implementation of this function using Intel AESNI.\r | |
| 510 | This function _aes_decrypt_cbc_hw SHOULD NOT be called directly. \r | |
| 511 | Developer should still call _aes_decrypt_cbc (in aes_modes_asm.s) which will poll cpu_capabilities and branch\r | |
| 512 | to this aesni-based function should it detecs that aesni is available.\r | |
| 513 | Blindly call this function SURELY will cause a CRASH on systems with no aesni support. \r | |
| 514 | \r | |
| 515 | Note that the decryption operation is not related over blocks.\r | |
| 516 | This gives opportunity of arranging aes_decrypt operations in parallel to speed up code.\r | |
| 517 | This is equivalent to what has been described in the Intel AES Instruction Set White Paper (Rev. 2.0 page 53-55)\r | |
| 518 | The following assembly code exploits this idea to achieve ~ 1.4 speed up in aes_decrypt_cbc.\r | |
| 519 | \r | |
| 520 | Example C code for packing 4 blocks in an iteration is shown as follows:\r | |
| 521 | \r | |
| 522 | while ((num_blk-=4)>=0) {\r | |
| 523 | \r | |
| 524 | // the following 4 functions can be interleaved to exploit parallelism\r | |
| 525 | aes_decrypt(ibuf, obuf, ctx);\r | |
| 526 | aes_decrypt(ibuf+1, obuf+1, ctx);\r | |
| 527 | aes_decrypt(ibuf+2, obuf+2, ctx);\r | |
| 528 | aes_decrypt(ibuf+3, obuf+3, ctx);\r | |
| 529 | \r | |
| 530 | obuf[0] ^= *iv; obuf[1] ^= ibuf[1]; obuf[2] ^= ibuf[1]; obuf[3] ^= ibuf[2];\r | |
| 531 | *iv = ibuf[3]; ibuf += 4; obuf += 4;\r | |
| 532 | }\r | |
| 533 | num_blk+=4;\r | |
| 534 | \r | |
| 535 | ----------------------------------------------------------------------------------------------------------------*/\r | |
| 536 | \r | |
| 537 | .text\r | |
| 538 | .align 4,0x90\r | |
| 539 | .globl _aes_decrypt_cbc_hw\r | |
| 540 | _aes_decrypt_cbc_hw:\r | |
| 541 | \r | |
| 542 | // push/save registers for local use\r | |
| 543 | #if defined __i386__\r | |
| 544 | \r | |
| 545 | push %ebp\r | |
| 546 | movl %esp, %ebp\r | |
| 547 | push %ebx // ibuf\r | |
| 548 | push %edi // obuf\r | |
| 549 | \r | |
| 550 | #define sp %esp\r | |
| 551 | \r | |
| 552 | #else // __x86_64__\r | |
| 553 | \r | |
| 554 | push %rbp\r | |
| 555 | mov %rsp, %rbp\r | |
| 556 | push %rbx\r | |
| 557 | push %r13\r | |
| 558 | push %r14\r | |
| 559 | push %r15\r | |
| 560 | \r | |
| 561 | #define sp %rsp\r | |
| 562 | \r | |
| 563 | #endif\r | |
| 564 | \r | |
| 565 | \r | |
| 566 | // if kernel, allocate stack space to save xmm registers\r | |
| 567 | #ifdef KERNEL\r | |
| 568 | #if defined __i386__\r | |
| 569 | sub $(8*16), %esp\r | |
| 570 | #else\r | |
| 571 | sub $(16*16), %rsp\r | |
| 572 | #endif\r | |
| 573 | movaps %xmm0, (sp)\r | |
| 574 | movaps %xmm1, 16(sp)\r | |
| 575 | movaps %xmm2, 32(sp)\r | |
| 576 | movaps %xmm3, 48(sp)\r | |
| 577 | movaps %xmm4, 64(sp)\r | |
| 578 | movaps %xmm5, 80(sp)\r | |
| 579 | movaps %xmm6, 96(sp)\r | |
| 580 | movaps %xmm7, 112(sp)\r | |
| 581 | #if defined __x86_64__\r | |
| 582 | movaps %xmm8, 16*8(sp)\r | |
| 583 | movaps %xmm9, 16*9(sp)\r | |
| 584 | movaps %xmm10, 16*10(sp)\r | |
| 585 | movaps %xmm11, 16*11(sp)\r | |
| 586 | movaps %xmm12, 16*12(sp)\r | |
| 587 | movaps %xmm13, 16*13(sp)\r | |
| 588 | movaps %xmm14, 16*14(sp)\r | |
| 589 | movaps %xmm15, 16*15(sp)\r | |
| 590 | #endif // __x86_64__\r | |
| 591 | #endif\r | |
| 592 | \r | |
| 593 | #undef iv\r | |
| 594 | #define iv %xmm0\r | |
| 595 | \r | |
| 596 | #if defined __i386__\r | |
| 597 | mov 12(%ebp), %eax // in_iv\r | |
| 598 | mov 24(%ebp), %edx // ctx\r | |
| 599 | movups (%eax), iv // iv = in_iv \r | |
| 600 | mov 8(%ebp), %ebx // ibuf\r | |
| 601 | mov 16(%ebp), %ecx // num_blk\r | |
| 602 | mov 20(%ebp), %edi // obuf\r | |
| 603 | \r | |
| 604 | #define ibuf %ebx\r | |
| 605 | #define obuf %edi\r | |
| 606 | #define num_blk %ecx \r | |
| 607 | #define ctx %edx\r | |
| 608 | \r | |
| 609 | #else // __x86_64__, rdi/rsi/rdx/rcx/r8\r | |
| 610 | \r | |
| 611 | mov %rdi, %rbx // ibuf\r | |
| 612 | movups (%rsi), iv // iv = in_iv\r | |
| 613 | mov %rdx, %r13 // num_blk\r | |
| 614 | mov %rcx, %r14 // obuf\r | |
| 615 | mov %r8, %r15 // ctx \r | |
| 616 | \r | |
| 617 | #define ibuf %rbx\r | |
| 618 | #define num_blk %r13d\r | |
| 619 | #define obuf %r14 \r | |
| 620 | #define ctx %r15\r | |
| 621 | \r | |
| 622 | #endif\r | |
| 623 | \r | |
| 624 | mov 240(ctx), %eax // aes length\r | |
| 625 | cmp $160, %eax // aes-128 decrypt\r | |
| 626 | je L_decrypt_128\r | |
| 627 | cmp $192, %eax // aes-192 decrypt\r | |
| 628 | je L_decrypt_192\r | |
| 629 | cmp $224, %eax // aes-256 decrypt\r | |
| 630 | je L_decrypt_256\r | |
| 631 | \r | |
| 632 | mov $-1, %eax // wrong aes length, to return -1\r | |
| 633 | jmp L_error // early exit due to wrong aes length\r | |
| 634 | \r | |
| 635 | \r | |
| 636 | //\r | |
| 637 | // aes-128 decrypt_cbc operation, after completion, branch to L_HW_cbc_done\r | |
| 638 | //\r | |
| 639 | \r | |
| 640 | L_decrypt_128:\r | |
| 641 | \r | |
| 642 | cmp $1, num_blk\r | |
| 643 | jl L_HW_cbc_done // if num_blk < 1, early return\r | |
| 644 | \r | |
| 645 | // aes-128 decrypt expanded keys\r | |
| 646 | movups 160(ctx), %xmm3\r | |
| 647 | movups 144(ctx), %xmm4\r | |
| 648 | movups 128(ctx), %xmm5\r | |
| 649 | movups 112(ctx), %xmm6\r | |
| 650 | movups 96(ctx), %xmm7\r | |
| 651 | #if defined __x86_64__\r | |
| 652 | movups 80(ctx), %xmm8\r | |
| 653 | movups 64(ctx), %xmm9\r | |
| 654 | movups 48(ctx), %xmm10\r | |
| 655 | movups 32(ctx), %xmm11\r | |
| 656 | movups 16(ctx), %xmm12\r | |
| 657 | movups 0(ctx), %xmm13\r | |
| 658 | #endif\r | |
| 659 | \r | |
| 660 | // performs 4 block decryption in an iteration to exploit decrypt in parallel\r | |
| 661 | \r | |
| 662 | // while ((num_blk-=4)>=0) {\r | |
| 663 | // aes_decrypt(ibuf, obuf, ctx);\r | |
| 664 | // aes_decrypt(ibuf+1, obuf+1, ctx);\r | |
| 665 | // aes_decrypt(ibuf+2, obuf+2, ctx);\r | |
| 666 | // aes_decrypt(ibuf+3, obuf+3, ctx);\r | |
| 667 | // obuf[0] ^= *iv; obuf[1] ^= ibuf[1]; obuf[2] ^= ibuf[1]; obuf[3] ^= ibuf[2];\r | |
| 668 | // *iv = ibuf[3]; ibuf += 4; obuf += 4;\r | |
| 669 | // }\r | |
| 670 | \r | |
| 671 | sub $4, num_blk // pre decrement num_blk by 4\r | |
| 672 | jl 9f // if num_blk < 4, skip the per-4-blocks processing code\r | |
| 673 | \r | |
| 674 | 0:\r | |
| 675 | \r | |
| 676 | \r | |
| 677 | #if defined __x86_64__\r | |
| 678 | \r | |
| 679 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r | |
| 680 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r | |
| 681 | movups 32(ibuf), %xmm14 // tmp = 3rd ibuf\r | |
| 682 | movups 48(ibuf), %xmm15 // tmp = 4th ibuf\r | |
| 683 | \r | |
| 684 | // for x86_64, the expanded keys are already stored in xmm3-xmm13\r | |
| 685 | \r | |
| 686 | // aes-128 decrypt round 0 per 4 blocks\r | |
| 687 | pxor %xmm3, %xmm1\r | |
| 688 | pxor %xmm3, %xmm2\r | |
| 689 | pxor %xmm3, %xmm14\r | |
| 690 | pxor %xmm3, %xmm15\r | |
| 691 | \r | |
| 692 | // aes-128 decrypt round 1 per 4 blocks\r | |
| 693 | aesdec %xmm4, %xmm1\r | |
| 694 | aesdec %xmm4, %xmm2\r | |
| 695 | aesdec %xmm4, %xmm14\r | |
| 696 | aesdec %xmm4, %xmm15\r | |
| 697 | \r | |
| 698 | // aes-128 decrypt round 2 per 4 blocks\r | |
| 699 | aesdec %xmm5, %xmm1\r | |
| 700 | aesdec %xmm5, %xmm2\r | |
| 701 | aesdec %xmm5, %xmm14\r | |
| 702 | aesdec %xmm5, %xmm15\r | |
| 703 | \r | |
| 704 | // aes-128 decrypt round 3 per 4 blocks\r | |
| 705 | aesdec %xmm6, %xmm1\r | |
| 706 | aesdec %xmm6, %xmm2\r | |
| 707 | aesdec %xmm6, %xmm14\r | |
| 708 | aesdec %xmm6, %xmm15\r | |
| 709 | \r | |
| 710 | // aes-128 decrypt round 4 per 4 blocks\r | |
| 711 | aesdec %xmm7, %xmm1\r | |
| 712 | aesdec %xmm7, %xmm2\r | |
| 713 | aesdec %xmm7, %xmm14\r | |
| 714 | aesdec %xmm7, %xmm15\r | |
| 715 | \r | |
| 716 | // aes-128 decrypt round 5 per 4 blocks\r | |
| 717 | aesdec %xmm8, %xmm1\r | |
| 718 | aesdec %xmm8, %xmm2\r | |
| 719 | aesdec %xmm8, %xmm14\r | |
| 720 | aesdec %xmm8, %xmm15\r | |
| 721 | \r | |
| 722 | // aes-128 decrypt round 6 per 4 blocks\r | |
| 723 | aesdec %xmm9, %xmm1\r | |
| 724 | aesdec %xmm9, %xmm2\r | |
| 725 | aesdec %xmm9, %xmm14\r | |
| 726 | aesdec %xmm9, %xmm15\r | |
| 727 | \r | |
| 728 | // aes-128 decrypt round 7 per 4 blocks\r | |
| 729 | aesdec %xmm10, %xmm1\r | |
| 730 | aesdec %xmm10, %xmm2\r | |
| 731 | aesdec %xmm10, %xmm14\r | |
| 732 | aesdec %xmm10, %xmm15\r | |
| 733 | \r | |
| 734 | // aes-128 decrypt round 8 per 4 blocks\r | |
| 735 | aesdec %xmm11, %xmm1\r | |
| 736 | aesdec %xmm11, %xmm2\r | |
| 737 | aesdec %xmm11, %xmm14\r | |
| 738 | aesdec %xmm11, %xmm15\r | |
| 739 | \r | |
| 740 | // aes-128 decrypt round 9 per 4 blocks\r | |
| 741 | aesdec %xmm12, %xmm1\r | |
| 742 | aesdec %xmm12, %xmm2\r | |
| 743 | aesdec %xmm12, %xmm14\r | |
| 744 | aesdec %xmm12, %xmm15\r | |
| 745 | \r | |
| 746 | // aes-128 decrypt round 10 (last) per 4 blocks\r | |
| 747 | aesdeclast %xmm13, %xmm1\r | |
| 748 | aesdeclast %xmm13, %xmm2\r | |
| 749 | aesdeclast %xmm13, %xmm14\r | |
| 750 | aesdeclast %xmm13, %xmm15\r | |
| 751 | \r | |
| 752 | pxor iv, %xmm1 // obuf[0] ^= *iv; \r | |
| 753 | movups (ibuf), iv // ibuf[0]\r | |
| 754 | pxor iv, %xmm2 // obuf[1] ^= ibuf[0]; \r | |
| 755 | movups 16(ibuf), iv // ibuf[1]\r | |
| 756 | pxor iv, %xmm14 // obuf[2] ^= ibuf[1]; \r | |
| 757 | movups 32(ibuf), iv // ibuf[2] \r | |
| 758 | pxor iv, %xmm15 // obuf[3] ^= obuf[2]; \r | |
| 759 | movups 48(ibuf), iv // *iv = ibuf[3]\r | |
| 760 | \r | |
| 761 | movups %xmm1, (obuf) // write 1st obuf\r | |
| 762 | movups %xmm2, 16(obuf) // write 2nd obuf\r | |
| 763 | movups %xmm14, 32(obuf) // write 3rd obuf\r | |
| 764 | movups %xmm15, 48(obuf) // write 4th obuf\r | |
| 765 | \r | |
| 766 | \r | |
| 767 | #else\r | |
| 768 | \r | |
| 769 | // aes_decrypt_cbc per 4 blocks using aes-128 for i386\r | |
| 770 | // xmm1/xmm2/xmm4/xmm5 used for obuf per block\r | |
| 771 | // xmm3 = key0\r | |
| 772 | // xmm0 = iv\r | |
| 773 | // xmm6/xmm7 dynamically load with other expanded keys\r | |
| 774 | \r | |
| 775 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r | |
| 776 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r | |
| 777 | movups 32(ibuf), %xmm4 // tmp = 3rd ibuf\r | |
| 778 | movups 48(ibuf), %xmm5 // tmp = 4th ibuf\r | |
| 779 | \r | |
| 780 | // aes_decrypt\r | |
| 781 | // for i386, sequentially load expanded keys into xmm6/xmm7\r | |
| 782 | \r | |
| 783 | movups 144(ctx), %xmm6 // key1\r | |
| 784 | \r | |
| 785 | // aes-128 decrypt round 0 per 4 blocks\r | |
| 786 | pxor %xmm3, %xmm1\r | |
| 787 | pxor %xmm3, %xmm2\r | |
| 788 | pxor %xmm3, %xmm4\r | |
| 789 | pxor %xmm3, %xmm5\r | |
| 790 | \r | |
| 791 | movups 128(ctx), %xmm7 // key2\r | |
| 792 | \r | |
| 793 | // aes-128 decrypt round 1 per 4 blocks\r | |
| 794 | aesdec %xmm6, %xmm1\r | |
| 795 | aesdec %xmm6, %xmm2\r | |
| 796 | aesdec %xmm6, %xmm4\r | |
| 797 | aesdec %xmm6, %xmm5\r | |
| 798 | \r | |
| 799 | movups 112(ctx), %xmm6 // key3\r | |
| 800 | \r | |
| 801 | // aes-128 decrypt round 2 per 4 blocks\r | |
| 802 | aesdec %xmm7, %xmm1\r | |
| 803 | aesdec %xmm7, %xmm2\r | |
| 804 | aesdec %xmm7, %xmm4\r | |
| 805 | aesdec %xmm7, %xmm5\r | |
| 806 | \r | |
| 807 | movups 96(ctx), %xmm7 // key4\r | |
| 808 | \r | |
| 809 | // aes-128 decrypt round 3 per 4 blocks\r | |
| 810 | aesdec %xmm6, %xmm1\r | |
| 811 | aesdec %xmm6, %xmm2\r | |
| 812 | aesdec %xmm6, %xmm4\r | |
| 813 | aesdec %xmm6, %xmm5\r | |
| 814 | \r | |
| 815 | movups 80(ctx), %xmm6 // key5\r | |
| 816 | \r | |
| 817 | // aes-128 decrypt round 4 per 4 blocks\r | |
| 818 | aesdec %xmm7, %xmm1\r | |
| 819 | aesdec %xmm7, %xmm2\r | |
| 820 | aesdec %xmm7, %xmm4\r | |
| 821 | aesdec %xmm7, %xmm5\r | |
| 822 | \r | |
| 823 | movups 64(ctx), %xmm7 // key6\r | |
| 824 | \r | |
| 825 | // aes-128 decrypt round 5 per 4 blocks\r | |
| 826 | aesdec %xmm6, %xmm1\r | |
| 827 | aesdec %xmm6, %xmm2\r | |
| 828 | aesdec %xmm6, %xmm4\r | |
| 829 | aesdec %xmm6, %xmm5\r | |
| 830 | \r | |
| 831 | movups 48(ctx), %xmm6 // key7\r | |
| 832 | \r | |
| 833 | // aes-128 decrypt round 6 per 4 blocks\r | |
| 834 | aesdec %xmm7, %xmm1\r | |
| 835 | aesdec %xmm7, %xmm2\r | |
| 836 | aesdec %xmm7, %xmm4\r | |
| 837 | aesdec %xmm7, %xmm5\r | |
| 838 | \r | |
| 839 | movups 32(ctx), %xmm7 // key8\r | |
| 840 | \r | |
| 841 | // aes-128 decrypt round 7 per 4 blocks\r | |
| 842 | aesdec %xmm6, %xmm1\r | |
| 843 | aesdec %xmm6, %xmm2\r | |
| 844 | aesdec %xmm6, %xmm4\r | |
| 845 | aesdec %xmm6, %xmm5\r | |
| 846 | \r | |
| 847 | movups 16(ctx), %xmm6 // key9\r | |
| 848 | \r | |
| 849 | // aes-128 decrypt round 8 per 4 blocks\r | |
| 850 | aesdec %xmm7, %xmm1\r | |
| 851 | aesdec %xmm7, %xmm2\r | |
| 852 | aesdec %xmm7, %xmm4\r | |
| 853 | aesdec %xmm7, %xmm5\r | |
| 854 | \r | |
| 855 | movups 0(ctx), %xmm7 // keyA\r | |
| 856 | \r | |
| 857 | // aes-128 decrypt round 9 per 4 blocks\r | |
| 858 | aesdec %xmm6, %xmm1\r | |
| 859 | aesdec %xmm6, %xmm2\r | |
| 860 | aesdec %xmm6, %xmm4\r | |
| 861 | aesdec %xmm6, %xmm5\r | |
| 862 | \r | |
| 863 | // aes-128 decrypt round 10 (last) per 4 blocks\r | |
| 864 | aesdeclast %xmm7, %xmm1\r | |
| 865 | aesdeclast %xmm7, %xmm2\r | |
| 866 | aesdeclast %xmm7, %xmm4\r | |
| 867 | aesdeclast %xmm7, %xmm5\r | |
| 868 | \r | |
| 869 | pxor iv, %xmm1 // 1st obuf ^= iv; \r | |
| 870 | movups (ibuf), iv // 1st memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 871 | pxor iv, %xmm2 // 2nd obuf ^= iv; \r | |
| 872 | movups 16(ibuf), iv // 2nd memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 873 | pxor iv, %xmm4 // 3rd obuf ^= iv; \r | |
| 874 | movups 32(ibuf), iv // 3rd memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 875 | pxor iv, %xmm5 // 4th obuf ^= iv; \r | |
| 876 | movups 48(ibuf), iv // 4th memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 877 | \r | |
| 878 | movups %xmm1, (obuf) // write 1st obuf\r | |
| 879 | movups %xmm2, 16(obuf) // write 2nd obuf\r | |
| 880 | movups %xmm4, 32(obuf) // write 3rd obuf\r | |
| 881 | movups %xmm5, 48(obuf) // write 4th obuf\r | |
| 882 | #endif\r | |
| 883 | \r | |
| 884 | add $64, ibuf // ibuf += 4; \r | |
| 885 | add $64, obuf // obuf += 4; \r | |
| 886 | \r | |
| 887 | sub $4, num_blk // num_blk -= 4\r | |
| 888 | jge 0b // if num_blk > 0, repeat the loop\r | |
| 889 | \r | |
| 890 | 9: add $4, num_blk // post incremtn num_blk by 4\r | |
| 891 | je L_HW_cbc_done // if num_blk == 0, no need for forthur processing code\r | |
| 892 | \r | |
| 893 | #if defined __i386__\r | |
| 894 | // updated as they might be needed as expanded keys in the remaining\r | |
| 895 | movups 144(ctx), %xmm4\r | |
| 896 | movups 128(ctx), %xmm5\r | |
| 897 | movups 112(ctx), %xmm6\r | |
| 898 | movups 96(ctx), %xmm7\r | |
| 899 | #endif\r | |
| 900 | \r | |
| 901 | test $2, num_blk // check whether num_blk has 2 blocks\r | |
| 902 | je 9f // if num_blk & 2 == 0, skip the per-pair processing code\r | |
| 903 | \r | |
| 904 | // do the remaining 2 blocks together\r | |
| 905 | \r | |
| 906 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r | |
| 907 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r | |
| 908 | \r | |
| 909 | // aes_decrypt\r | |
| 910 | pxor %xmm3, %xmm1\r | |
| 911 | pxor %xmm3, %xmm2\r | |
| 912 | aesdec %xmm4, %xmm1\r | |
| 913 | aesdec %xmm4, %xmm2\r | |
| 914 | aesdec %xmm5, %xmm1\r | |
| 915 | aesdec %xmm5, %xmm2\r | |
| 916 | aesdec %xmm6, %xmm1\r | |
| 917 | aesdec %xmm6, %xmm2\r | |
| 918 | #if defined __x86_64__\r | |
| 919 | aesdec %xmm7, %xmm1\r | |
| 920 | aesdec %xmm7, %xmm2\r | |
| 921 | aesdec %xmm8, %xmm1\r | |
| 922 | aesdec %xmm8, %xmm2\r | |
| 923 | aesdec %xmm9, %xmm1\r | |
| 924 | aesdec %xmm9, %xmm2\r | |
| 925 | aesdec %xmm10, %xmm1\r | |
| 926 | aesdec %xmm10, %xmm2\r | |
| 927 | aesdec %xmm11, %xmm1\r | |
| 928 | aesdec %xmm11, %xmm2\r | |
| 929 | aesdec %xmm12, %xmm1\r | |
| 930 | aesdec %xmm12, %xmm2\r | |
| 931 | aesdeclast %xmm13, %xmm1\r | |
| 932 | aesdeclast %xmm13, %xmm2\r | |
| 933 | #else\r | |
| 934 | movups 80(ctx), %xmm6\r | |
| 935 | aesdec %xmm7, %xmm1\r | |
| 936 | aesdec %xmm7, %xmm2\r | |
| 937 | movups 64(ctx), %xmm7\r | |
| 938 | aesdec %xmm6, %xmm1\r | |
| 939 | aesdec %xmm6, %xmm2\r | |
| 940 | movups 48(ctx), %xmm6\r | |
| 941 | aesdec %xmm7, %xmm1\r | |
| 942 | aesdec %xmm7, %xmm2\r | |
| 943 | movups 32(ctx), %xmm7\r | |
| 944 | aesdec %xmm6, %xmm1\r | |
| 945 | aesdec %xmm6, %xmm2\r | |
| 946 | movups 16(ctx), %xmm6\r | |
| 947 | aesdec %xmm7, %xmm1\r | |
| 948 | aesdec %xmm7, %xmm2\r | |
| 949 | movups 0(ctx), %xmm7\r | |
| 950 | aesdec %xmm6, %xmm1\r | |
| 951 | aesdec %xmm6, %xmm2\r | |
| 952 | aesdeclast %xmm7, %xmm1\r | |
| 953 | aesdeclast %xmm7, %xmm2\r | |
| 954 | movups 112(ctx), %xmm6\r | |
| 955 | movups 96(ctx), %xmm7\r | |
| 956 | #endif\r | |
| 957 | \r | |
| 958 | pxor iv, %xmm1 // obuf[0] ^= *iv; \r | |
| 959 | movups (ibuf), iv // ibuf[0]\r | |
| 960 | pxor iv, %xmm2 // obuf[1] ^= ibuf[0]\r | |
| 961 | movups 16(ibuf), iv // *iv = ibuf[1]\r | |
| 962 | \r | |
| 963 | movups %xmm1, (obuf) // write obuf[0]\r | |
| 964 | movups %xmm2, 16(obuf) // write obuf[1]\r | |
| 965 | \r | |
| 966 | add $32, ibuf // ibuf += 2\r | |
| 967 | add $32, obuf // obuf += 2\r | |
| 968 | \r | |
| 969 | 9:\r | |
| 970 | test $1, num_blk // check whether num_blk has residual 1 block\r | |
| 971 | je L_HW_cbc_done // if num_blk == 0, no need for residual processing code\r | |
| 972 | \r | |
| 973 | movups (ibuf), %xmm2 // tmp = ibuf\r | |
| 974 | // aes_decrypt\r | |
| 975 | pxor %xmm3, %xmm2\r | |
| 976 | aesdec %xmm4, %xmm2\r | |
| 977 | aesdec %xmm5, %xmm2\r | |
| 978 | aesdec %xmm6, %xmm2\r | |
| 979 | aesdec %xmm7, %xmm2\r | |
| 980 | #if defined __x86_64__\r | |
| 981 | aesdec %xmm8, %xmm2\r | |
| 982 | aesdec %xmm9, %xmm2\r | |
| 983 | aesdec %xmm10, %xmm2\r | |
| 984 | aesdec %xmm11, %xmm2\r | |
| 985 | aesdec %xmm12, %xmm2\r | |
| 986 | aesdeclast %xmm13, %xmm2\r | |
| 987 | #else\r | |
| 988 | movups 80(ctx), %xmm1\r | |
| 989 | aesdec %xmm1, %xmm2\r | |
| 990 | movups 64(ctx), %xmm1\r | |
| 991 | aesdec %xmm1, %xmm2\r | |
| 992 | movups 48(ctx), %xmm1\r | |
| 993 | aesdec %xmm1, %xmm2\r | |
| 994 | movups 32(ctx), %xmm1\r | |
| 995 | aesdec %xmm1, %xmm2\r | |
| 996 | movups 16(ctx), %xmm1\r | |
| 997 | aesdec %xmm1, %xmm2\r | |
| 998 | movups (ctx), %xmm1\r | |
| 999 | aesdeclast %xmm1, %xmm2\r | |
| 1000 | #endif\r | |
| 1001 | \r | |
| 1002 | pxor iv, %xmm2 // *obuf ^= *iv; \r | |
| 1003 | movups (ibuf), iv // *iv = *ibuf;\r | |
| 1004 | movups %xmm2, (obuf) // write *obuf\r | |
| 1005 | \r | |
| 1006 | jmp L_HW_cbc_done\r | |
| 1007 | \r | |
| 1008 | //\r | |
| 1009 | // aes-192 decrypt_cbc operation, after completion, branch to L_HW_cbc_done\r | |
| 1010 | //\r | |
| 1011 | \r | |
| 1012 | L_decrypt_192:\r | |
| 1013 | \r | |
| 1014 | cmp $1, num_blk\r | |
| 1015 | jl L_HW_cbc_done // if num_blk < 1, early return\r | |
| 1016 | \r | |
| 1017 | // aes-192 decryp expanded keys\r | |
| 1018 | movups 192(ctx), %xmm3\r | |
| 1019 | movups 176(ctx), %xmm4\r | |
| 1020 | movups 160(ctx), %xmm5\r | |
| 1021 | movups 144(ctx), %xmm6\r | |
| 1022 | movups 128(ctx), %xmm7\r | |
| 1023 | #if defined __x86_64__\r | |
| 1024 | movups 112(ctx), %xmm8\r | |
| 1025 | movups 96(ctx), %xmm9\r | |
| 1026 | movups 80(ctx), %xmm10\r | |
| 1027 | movups 64(ctx), %xmm11\r | |
| 1028 | movups 48(ctx), %xmm12\r | |
| 1029 | movups 32(ctx), %xmm13\r | |
| 1030 | movups 16(ctx), %xmm14\r | |
| 1031 | movups (ctx), %xmm15\r | |
| 1032 | #endif\r | |
| 1033 | \r | |
| 1034 | // performs 4 block decryption in an iteration to exploit decrypt in parallel\r | |
| 1035 | \r | |
| 1036 | // while ((num_blk-=4)>=0) {\r | |
| 1037 | // aes_decrypt(ibuf, obuf, ctx);\r | |
| 1038 | // aes_decrypt(ibuf+1, obuf+1, ctx);\r | |
| 1039 | // aes_decrypt(ibuf+2, obuf+2, ctx);\r | |
| 1040 | // aes_decrypt(ibuf+3, obuf+3, ctx);\r | |
| 1041 | // obuf[0] ^= *iv; obuf[1] ^= ibuf[1]; obuf[2] ^= ibuf[1]; obuf[3] ^= ibuf[2];\r | |
| 1042 | // *iv = ibuf[3]; ibuf += 4; obuf += 4;\r | |
| 1043 | // }\r | |
| 1044 | \r | |
| 1045 | sub $4, num_blk // pre decrement num_blk by 4\r | |
| 1046 | jl 9f // if num_blk < 4, skip the per-4-blocks processing code\r | |
| 1047 | 0:\r | |
| 1048 | \r | |
| 1049 | #if defined __x86_64__\r | |
| 1050 | \r | |
| 1051 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r | |
| 1052 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r | |
| 1053 | movups 32(ibuf), %xmm14 // tmp = 3rd ibuf\r | |
| 1054 | movups 48(ibuf), %xmm15 // tmp = 4th ibuf\r | |
| 1055 | \r | |
| 1056 | // aes_decrypt, for x86_64, the expanded keys are already stored in xmm3-xmm13\r | |
| 1057 | // use %xmm12/%xmm13 ts dynamic keys in the middle, restored afterwards\r | |
| 1058 | \r | |
| 1059 | // round 0 for 4 blocks\r | |
| 1060 | pxor %xmm3, %xmm1\r | |
| 1061 | pxor %xmm3, %xmm2\r | |
| 1062 | pxor %xmm3, %xmm14\r | |
| 1063 | pxor %xmm3, %xmm15\r | |
| 1064 | \r | |
| 1065 | // round 1 for 4 blocks\r | |
| 1066 | aesdec %xmm4, %xmm1\r | |
| 1067 | aesdec %xmm4, %xmm2\r | |
| 1068 | aesdec %xmm4, %xmm14\r | |
| 1069 | aesdec %xmm4, %xmm15\r | |
| 1070 | \r | |
| 1071 | // round 2 for 4 blocks\r | |
| 1072 | aesdec %xmm5, %xmm1\r | |
| 1073 | aesdec %xmm5, %xmm2\r | |
| 1074 | aesdec %xmm5, %xmm14\r | |
| 1075 | aesdec %xmm5, %xmm15\r | |
| 1076 | \r | |
| 1077 | // round 3 for 4 blocks\r | |
| 1078 | aesdec %xmm6, %xmm1\r | |
| 1079 | aesdec %xmm6, %xmm2\r | |
| 1080 | aesdec %xmm6, %xmm14\r | |
| 1081 | aesdec %xmm6, %xmm15\r | |
| 1082 | \r | |
| 1083 | // round 4 for 4 blocks\r | |
| 1084 | aesdec %xmm7, %xmm1\r | |
| 1085 | aesdec %xmm7, %xmm2\r | |
| 1086 | aesdec %xmm7, %xmm14\r | |
| 1087 | aesdec %xmm7, %xmm15\r | |
| 1088 | \r | |
| 1089 | // round 5 for 4 blocks\r | |
| 1090 | aesdec %xmm8, %xmm1\r | |
| 1091 | aesdec %xmm8, %xmm2\r | |
| 1092 | aesdec %xmm8, %xmm14\r | |
| 1093 | aesdec %xmm8, %xmm15\r | |
| 1094 | \r | |
| 1095 | // round 6 for 4 blocks\r | |
| 1096 | aesdec %xmm9, %xmm1\r | |
| 1097 | aesdec %xmm9, %xmm2\r | |
| 1098 | aesdec %xmm9, %xmm14\r | |
| 1099 | aesdec %xmm9, %xmm15\r | |
| 1100 | \r | |
| 1101 | // round 7 for 4 blocks\r | |
| 1102 | aesdec %xmm10, %xmm1\r | |
| 1103 | aesdec %xmm10, %xmm2\r | |
| 1104 | aesdec %xmm10, %xmm14\r | |
| 1105 | aesdec %xmm10, %xmm15\r | |
| 1106 | \r | |
| 1107 | // round 8 for 4 blocks\r | |
| 1108 | aesdec %xmm11, %xmm1\r | |
| 1109 | aesdec %xmm11, %xmm2\r | |
| 1110 | aesdec %xmm11, %xmm14\r | |
| 1111 | aesdec %xmm11, %xmm15\r | |
| 1112 | \r | |
| 1113 | // round 9 for 4 blocks\r | |
| 1114 | aesdec %xmm12, %xmm1\r | |
| 1115 | aesdec %xmm12, %xmm2\r | |
| 1116 | aesdec %xmm12, %xmm14\r | |
| 1117 | aesdec %xmm12, %xmm15\r | |
| 1118 | \r | |
| 1119 | movups 16(ctx), %xmm12\r | |
| 1120 | \r | |
| 1121 | // round A for 4 blocks\r | |
| 1122 | aesdec %xmm13, %xmm1\r | |
| 1123 | aesdec %xmm13, %xmm2\r | |
| 1124 | aesdec %xmm13, %xmm14\r | |
| 1125 | aesdec %xmm13, %xmm15\r | |
| 1126 | \r | |
| 1127 | movups (ctx), %xmm13\r | |
| 1128 | \r | |
| 1129 | // round B for 4 blocks\r | |
| 1130 | aesdec %xmm12, %xmm1\r | |
| 1131 | aesdec %xmm12, %xmm2\r | |
| 1132 | aesdec %xmm12, %xmm14\r | |
| 1133 | aesdec %xmm12, %xmm15\r | |
| 1134 | \r | |
| 1135 | movups 48(ctx), %xmm12 // restore %xmm12 to its original key\r | |
| 1136 | \r | |
| 1137 | // round C (last) for 4 blocks\r | |
| 1138 | aesdeclast %xmm13, %xmm1\r | |
| 1139 | aesdeclast %xmm13, %xmm2\r | |
| 1140 | aesdeclast %xmm13, %xmm14\r | |
| 1141 | aesdeclast %xmm13, %xmm15\r | |
| 1142 | \r | |
| 1143 | movups 32(ctx), %xmm13 // restore %xmm13 to its original key\r | |
| 1144 | \r | |
| 1145 | pxor iv, %xmm1 // obuf[0] ^= *iv; \r | |
| 1146 | movups (ibuf), iv // ibuf[0]\r | |
| 1147 | pxor iv, %xmm2 // obuf[1] ^= ibuf[0] \r | |
| 1148 | movups 16(ibuf), iv // ibuf[1]\r | |
| 1149 | pxor iv, %xmm14 // obuf[2] ^= ibuf[1] \r | |
| 1150 | movups 32(ibuf), iv // ibuf[2] \r | |
| 1151 | pxor iv, %xmm15 // obuf[3] ^= ibuf[2] \r | |
| 1152 | movups 48(ibuf), iv // *iv = ibuf[3] \r | |
| 1153 | \r | |
| 1154 | movups %xmm1, (obuf) // write 1st obuf\r | |
| 1155 | movups %xmm2, 16(obuf) // write 2nd obuf\r | |
| 1156 | movups %xmm14, 32(obuf) // write 3rd obuf\r | |
| 1157 | movups %xmm15, 48(obuf) // write 4th obuf\r | |
| 1158 | \r | |
| 1159 | add $64, ibuf // ibuf += 4; \r | |
| 1160 | add $64, obuf // obuf += 4; \r | |
| 1161 | \r | |
| 1162 | sub $4, num_blk // num_blk -= 4\r | |
| 1163 | jge 0b // if num_blk > 0, repeat the loop\r | |
| 1164 | \r | |
| 1165 | 9: add $4, num_blk // post incremtn num_blk by 4\r | |
| 1166 | je L_HW_cbc_done // if num_blk == 0, prepare to return \r | |
| 1167 | \r | |
| 1168 | movups 16(ctx), %xmm14 // restore %xmm14 to its key\r | |
| 1169 | movups (ctx), %xmm15 // restore %xmm15 to its key\r | |
| 1170 | \r | |
| 1171 | #else\r | |
| 1172 | \r | |
| 1173 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r | |
| 1174 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r | |
| 1175 | movups 32(ibuf), %xmm4 // tmp = 3rd ibuf\r | |
| 1176 | movups 48(ibuf), %xmm5 // tmp = 4th ibuf\r | |
| 1177 | \r | |
| 1178 | // aes_decrypt\r | |
| 1179 | // for i386, sequentially load expanded keys into xmm6/xmm7\r | |
| 1180 | movups 176(ctx), %xmm6\r | |
| 1181 | pxor %xmm3, %xmm1\r | |
| 1182 | pxor %xmm3, %xmm2\r | |
| 1183 | pxor %xmm3, %xmm4\r | |
| 1184 | pxor %xmm3, %xmm5\r | |
| 1185 | \r | |
| 1186 | movups 160(ctx), %xmm7\r | |
| 1187 | aesdec %xmm6, %xmm1\r | |
| 1188 | aesdec %xmm6, %xmm2\r | |
| 1189 | aesdec %xmm6, %xmm4\r | |
| 1190 | aesdec %xmm6, %xmm5\r | |
| 1191 | \r | |
| 1192 | movups 144(ctx), %xmm6\r | |
| 1193 | aesdec %xmm7, %xmm1\r | |
| 1194 | aesdec %xmm7, %xmm2\r | |
| 1195 | aesdec %xmm7, %xmm4\r | |
| 1196 | aesdec %xmm7, %xmm5\r | |
| 1197 | \r | |
| 1198 | movups 128(ctx), %xmm7\r | |
| 1199 | aesdec %xmm6, %xmm1\r | |
| 1200 | aesdec %xmm6, %xmm2\r | |
| 1201 | aesdec %xmm6, %xmm4\r | |
| 1202 | aesdec %xmm6, %xmm5\r | |
| 1203 | \r | |
| 1204 | movups 112(ctx), %xmm6\r | |
| 1205 | aesdec %xmm7, %xmm1\r | |
| 1206 | aesdec %xmm7, %xmm2\r | |
| 1207 | aesdec %xmm7, %xmm4\r | |
| 1208 | aesdec %xmm7, %xmm5\r | |
| 1209 | \r | |
| 1210 | movups 96(ctx), %xmm7\r | |
| 1211 | aesdec %xmm6, %xmm1\r | |
| 1212 | aesdec %xmm6, %xmm2\r | |
| 1213 | aesdec %xmm6, %xmm4\r | |
| 1214 | aesdec %xmm6, %xmm5\r | |
| 1215 | \r | |
| 1216 | movups 80(ctx), %xmm6\r | |
| 1217 | aesdec %xmm7, %xmm1\r | |
| 1218 | aesdec %xmm7, %xmm2\r | |
| 1219 | aesdec %xmm7, %xmm4\r | |
| 1220 | aesdec %xmm7, %xmm5\r | |
| 1221 | \r | |
| 1222 | movups 64(ctx), %xmm7\r | |
| 1223 | aesdec %xmm6, %xmm1\r | |
| 1224 | aesdec %xmm6, %xmm2\r | |
| 1225 | aesdec %xmm6, %xmm4\r | |
| 1226 | aesdec %xmm6, %xmm5\r | |
| 1227 | \r | |
| 1228 | movups 48(ctx), %xmm6\r | |
| 1229 | aesdec %xmm7, %xmm1\r | |
| 1230 | aesdec %xmm7, %xmm2\r | |
| 1231 | aesdec %xmm7, %xmm4\r | |
| 1232 | aesdec %xmm7, %xmm5\r | |
| 1233 | \r | |
| 1234 | movups 32(ctx), %xmm7\r | |
| 1235 | aesdec %xmm6, %xmm1\r | |
| 1236 | aesdec %xmm6, %xmm2\r | |
| 1237 | aesdec %xmm6, %xmm4\r | |
| 1238 | aesdec %xmm6, %xmm5\r | |
| 1239 | \r | |
| 1240 | movups 16(ctx), %xmm6\r | |
| 1241 | aesdec %xmm7, %xmm1\r | |
| 1242 | aesdec %xmm7, %xmm2\r | |
| 1243 | aesdec %xmm7, %xmm4\r | |
| 1244 | aesdec %xmm7, %xmm5\r | |
| 1245 | \r | |
| 1246 | movups 0(ctx), %xmm7\r | |
| 1247 | aesdec %xmm6, %xmm1\r | |
| 1248 | aesdec %xmm6, %xmm2\r | |
| 1249 | aesdec %xmm6, %xmm4\r | |
| 1250 | aesdec %xmm6, %xmm5\r | |
| 1251 | \r | |
| 1252 | aesdeclast %xmm7, %xmm1\r | |
| 1253 | aesdeclast %xmm7, %xmm2\r | |
| 1254 | aesdeclast %xmm7, %xmm4\r | |
| 1255 | aesdeclast %xmm7, %xmm5\r | |
| 1256 | \r | |
| 1257 | pxor iv, %xmm1 // 1st obuf ^= iv; \r | |
| 1258 | movups (ibuf), iv // 1st memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1259 | pxor iv, %xmm2 // 2nd obuf ^= iv; \r | |
| 1260 | movups 16(ibuf), iv // 2nd memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1261 | pxor iv, %xmm4 // 3rd obuf ^= iv; \r | |
| 1262 | movups 32(ibuf), iv // 3rd memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1263 | pxor iv, %xmm5 // 4th obuf ^= iv; \r | |
| 1264 | movups 48(ibuf), iv // 4th memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1265 | movups %xmm1, (obuf) // write 1st obuf\r | |
| 1266 | movups %xmm2, 16(obuf) // write 2nd obuf\r | |
| 1267 | movups %xmm4, 32(obuf) // write 3rd obuf\r | |
| 1268 | movups %xmm5, 48(obuf) // write 4th obuf\r | |
| 1269 | \r | |
| 1270 | add $64, ibuf // ibuf += AES_BLOCK_SIZE * 4; \r | |
| 1271 | add $64, obuf // obuf += AES_BLOCK_SIZE * 4; \r | |
| 1272 | \r | |
| 1273 | sub $4, num_blk // num_blk -= 4\r | |
| 1274 | jge 0b // if num_blk > 0, repeat the loop\r | |
| 1275 | \r | |
| 1276 | \r | |
| 1277 | 9: add $4, num_blk // post incremtn num_blk by 4\r | |
| 1278 | je L_HW_cbc_done // if num_blk == 0, no need for forthur processing code\r | |
| 1279 | \r | |
| 1280 | movups 176(ctx), %xmm4\r | |
| 1281 | movups 160(ctx), %xmm5\r | |
| 1282 | movups 144(ctx), %xmm6\r | |
| 1283 | movups 128(ctx), %xmm7\r | |
| 1284 | \r | |
| 1285 | #endif\r | |
| 1286 | \r | |
| 1287 | // per-block aes_decrypt_cbc loop\r | |
| 1288 | \r | |
| 1289 | 0:\r | |
| 1290 | movups (ibuf), %xmm2 // tmp = ibuf\r | |
| 1291 | \r | |
| 1292 | // aes_decrypt\r | |
| 1293 | pxor %xmm3, %xmm2\r | |
| 1294 | aesdec %xmm4, %xmm2\r | |
| 1295 | aesdec %xmm5, %xmm2\r | |
| 1296 | aesdec %xmm6, %xmm2\r | |
| 1297 | aesdec %xmm7, %xmm2\r | |
| 1298 | #if defined __x86_64__\r | |
| 1299 | aesdec %xmm8, %xmm2\r | |
| 1300 | aesdec %xmm9, %xmm2\r | |
| 1301 | aesdec %xmm10, %xmm2\r | |
| 1302 | aesdec %xmm11, %xmm2\r | |
| 1303 | aesdec %xmm12, %xmm2\r | |
| 1304 | aesdec %xmm13, %xmm2\r | |
| 1305 | aesdec %xmm14, %xmm2\r | |
| 1306 | aesdeclast %xmm15, %xmm2\r | |
| 1307 | #else\r | |
| 1308 | movups 112(ctx), %xmm1\r | |
| 1309 | aesdec %xmm1, %xmm2\r | |
| 1310 | movups 96(ctx), %xmm1\r | |
| 1311 | aesdec %xmm1, %xmm2\r | |
| 1312 | movups 80(ctx), %xmm1\r | |
| 1313 | aesdec %xmm1, %xmm2\r | |
| 1314 | movups 64(ctx), %xmm1\r | |
| 1315 | aesdec %xmm1, %xmm2\r | |
| 1316 | movups 48(ctx), %xmm1\r | |
| 1317 | aesdec %xmm1, %xmm2\r | |
| 1318 | movups 32(ctx), %xmm1\r | |
| 1319 | aesdec %xmm1, %xmm2\r | |
| 1320 | movups 16(ctx), %xmm1\r | |
| 1321 | aesdec %xmm1, %xmm2\r | |
| 1322 | movups (ctx), %xmm1\r | |
| 1323 | aesdeclast %xmm1, %xmm2\r | |
| 1324 | #endif\r | |
| 1325 | \r | |
| 1326 | pxor iv, %xmm2 // obuf ^= iv; \r | |
| 1327 | movups (ibuf), iv // memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1328 | \r | |
| 1329 | movups %xmm2, (obuf) // write obuf\r | |
| 1330 | \r | |
| 1331 | add $16, ibuf // ibuf += AES_BLOCK_SIZE; \r | |
| 1332 | add $16, obuf // obuf += AES_BLOCK_SIZE; \r | |
| 1333 | sub $1, num_blk // num_blk --\r | |
| 1334 | jg 0b // if num_blk > 0, repeat the loop\r | |
| 1335 | \r | |
| 1336 | jmp L_HW_cbc_done\r | |
| 1337 | \r | |
| 1338 | //\r | |
| 1339 | // aes-256 decrypt_cbc operation, after completion, branch to L_HW_cbc_done\r | |
| 1340 | //\r | |
| 1341 | \r | |
| 1342 | L_decrypt_256:\r | |
| 1343 | \r | |
| 1344 | cmp $1, num_blk\r | |
| 1345 | jl L_HW_cbc_done \r | |
| 1346 | \r | |
| 1347 | movups 224(ctx), %xmm3\r | |
| 1348 | movups 208(ctx), %xmm4\r | |
| 1349 | movups 192(ctx), %xmm5\r | |
| 1350 | movups 176(ctx), %xmm6\r | |
| 1351 | movups 160(ctx), %xmm7\r | |
| 1352 | #if defined __x86_64__\r | |
| 1353 | movups 144(ctx), %xmm8\r | |
| 1354 | movups 128(ctx), %xmm9\r | |
| 1355 | movups 112(ctx), %xmm10\r | |
| 1356 | movups 96(ctx), %xmm11\r | |
| 1357 | movups 80(ctx), %xmm12\r | |
| 1358 | movups 64(ctx), %xmm13\r | |
| 1359 | movups 48(ctx), %xmm14\r | |
| 1360 | movups 32(ctx), %xmm15\r | |
| 1361 | // movups 16(ctx), %xmm14\r | |
| 1362 | // movups (ctx), %xmm15\r | |
| 1363 | #endif\r | |
| 1364 | \r | |
| 1365 | #if defined __x86_64__\r | |
| 1366 | \r | |
| 1367 | sub $4, num_blk // pre decrement num_blk by 4\r | |
| 1368 | jl 9f // if num_blk < 4, skip the per-4-blocks processing code\r | |
| 1369 | 0:\r | |
| 1370 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r | |
| 1371 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r | |
| 1372 | movups 32(ibuf), %xmm14 // tmp = 3rd ibuf\r | |
| 1373 | movups 48(ibuf), %xmm15 // tmp = 4th ibuf\r | |
| 1374 | \r | |
| 1375 | // aes_decrypt, for x86_64, the expanded keys are already stored in xmm3-xmm13\r | |
| 1376 | pxor %xmm3, %xmm1\r | |
| 1377 | pxor %xmm3, %xmm2\r | |
| 1378 | pxor %xmm3, %xmm14\r | |
| 1379 | pxor %xmm3, %xmm15\r | |
| 1380 | \r | |
| 1381 | aesdec %xmm4, %xmm1\r | |
| 1382 | aesdec %xmm4, %xmm2\r | |
| 1383 | aesdec %xmm4, %xmm14\r | |
| 1384 | aesdec %xmm4, %xmm15\r | |
| 1385 | \r | |
| 1386 | aesdec %xmm5, %xmm1\r | |
| 1387 | aesdec %xmm5, %xmm2\r | |
| 1388 | aesdec %xmm5, %xmm14\r | |
| 1389 | aesdec %xmm5, %xmm15\r | |
| 1390 | \r | |
| 1391 | aesdec %xmm6, %xmm1\r | |
| 1392 | aesdec %xmm6, %xmm2\r | |
| 1393 | aesdec %xmm6, %xmm14\r | |
| 1394 | aesdec %xmm6, %xmm15\r | |
| 1395 | \r | |
| 1396 | aesdec %xmm7, %xmm1\r | |
| 1397 | aesdec %xmm7, %xmm2\r | |
| 1398 | aesdec %xmm7, %xmm14\r | |
| 1399 | aesdec %xmm7, %xmm15\r | |
| 1400 | \r | |
| 1401 | aesdec %xmm8, %xmm1\r | |
| 1402 | aesdec %xmm8, %xmm2\r | |
| 1403 | aesdec %xmm8, %xmm14\r | |
| 1404 | aesdec %xmm8, %xmm15\r | |
| 1405 | \r | |
| 1406 | aesdec %xmm9, %xmm1\r | |
| 1407 | aesdec %xmm9, %xmm2\r | |
| 1408 | aesdec %xmm9, %xmm14\r | |
| 1409 | aesdec %xmm9, %xmm15\r | |
| 1410 | \r | |
| 1411 | aesdec %xmm10, %xmm1\r | |
| 1412 | aesdec %xmm10, %xmm2\r | |
| 1413 | aesdec %xmm10, %xmm14\r | |
| 1414 | aesdec %xmm10, %xmm15\r | |
| 1415 | \r | |
| 1416 | aesdec %xmm11, %xmm1\r | |
| 1417 | aesdec %xmm11, %xmm2\r | |
| 1418 | aesdec %xmm11, %xmm14\r | |
| 1419 | aesdec %xmm11, %xmm15\r | |
| 1420 | \r | |
| 1421 | aesdec %xmm12, %xmm1\r | |
| 1422 | aesdec %xmm12, %xmm2\r | |
| 1423 | aesdec %xmm12, %xmm14\r | |
| 1424 | aesdec %xmm12, %xmm15\r | |
| 1425 | movups 48(ctx), %xmm12\r | |
| 1426 | \r | |
| 1427 | aesdec %xmm13, %xmm1\r | |
| 1428 | aesdec %xmm13, %xmm2\r | |
| 1429 | aesdec %xmm13, %xmm14\r | |
| 1430 | aesdec %xmm13, %xmm15\r | |
| 1431 | movups 32(ctx), %xmm13\r | |
| 1432 | \r | |
| 1433 | aesdec %xmm12, %xmm1\r | |
| 1434 | aesdec %xmm12, %xmm2\r | |
| 1435 | aesdec %xmm12, %xmm14\r | |
| 1436 | aesdec %xmm12, %xmm15\r | |
| 1437 | movups 16(ctx), %xmm12\r | |
| 1438 | \r | |
| 1439 | aesdec %xmm13, %xmm1\r | |
| 1440 | aesdec %xmm13, %xmm2\r | |
| 1441 | aesdec %xmm13, %xmm14\r | |
| 1442 | aesdec %xmm13, %xmm15\r | |
| 1443 | movups (ctx), %xmm13\r | |
| 1444 | \r | |
| 1445 | aesdec %xmm12, %xmm1\r | |
| 1446 | aesdec %xmm12, %xmm2\r | |
| 1447 | aesdec %xmm12, %xmm14\r | |
| 1448 | aesdec %xmm12, %xmm15\r | |
| 1449 | movups 80(ctx), %xmm12\r | |
| 1450 | \r | |
| 1451 | aesdeclast %xmm13, %xmm1\r | |
| 1452 | aesdeclast %xmm13, %xmm2\r | |
| 1453 | aesdeclast %xmm13, %xmm14\r | |
| 1454 | aesdeclast %xmm13, %xmm15\r | |
| 1455 | movups 64(ctx), %xmm13\r | |
| 1456 | \r | |
| 1457 | pxor iv, %xmm1 // obuf ^= iv; \r | |
| 1458 | movups (ibuf), iv // memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1459 | pxor iv, %xmm2 // obuf ^= iv; \r | |
| 1460 | movups 16(ibuf), iv // memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1461 | pxor iv, %xmm14 // obuf ^= iv; \r | |
| 1462 | movups 32(ibuf), iv // memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1463 | pxor iv, %xmm15 // obuf ^= iv; \r | |
| 1464 | movups 48(ibuf), iv // memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1465 | \r | |
| 1466 | movups %xmm1, (obuf) // write 1st obuf\r | |
| 1467 | movups %xmm2, 16(obuf) // write 2nd obuf\r | |
| 1468 | movups %xmm14, 32(obuf) // write 3rd obuf\r | |
| 1469 | movups %xmm15, 48(obuf) // write 4th obuf\r | |
| 1470 | \r | |
| 1471 | add $64, ibuf // ibuf += AES_BLOCK_SIZE*4; \r | |
| 1472 | add $64, obuf // obuf += AES_BLOCK_SIZE*4; \r | |
| 1473 | \r | |
| 1474 | sub $4, num_blk // num_blk -= 4\r | |
| 1475 | jge 0b // if num_blk > 0, repeat the loop\r | |
| 1476 | \r | |
| 1477 | 9: add $4, num_blk // post incremtn num_blk by 4\r | |
| 1478 | je L_HW_cbc_done // if num_blk == 0, no need for forthur processing code\r | |
| 1479 | \r | |
| 1480 | movups 48(ctx), %xmm14\r | |
| 1481 | movups 32(ctx), %xmm15\r | |
| 1482 | \r | |
| 1483 | #else\r | |
| 1484 | \r | |
| 1485 | sub $4, num_blk // pre decrement num_blk by 4\r | |
| 1486 | jl 9f // if num_blk < 4, skip the per-pair processing code\r | |
| 1487 | 0:\r | |
| 1488 | movups (ibuf), %xmm1 // tmp = 1st ibuf\r | |
| 1489 | movups 16(ibuf), %xmm2 // tmp = 2nd ibuf\r | |
| 1490 | movups 32(ibuf), %xmm4 // tmp = 3rd ibuf\r | |
| 1491 | movups 48(ibuf), %xmm5 // tmp = 4th ibuf\r | |
| 1492 | \r | |
| 1493 | // aes_decrypt\r | |
| 1494 | // for i386, sequentially load expanded keys into xmm6/xmm7\r | |
| 1495 | movups 208(ctx), %xmm6\r | |
| 1496 | pxor %xmm3, %xmm1\r | |
| 1497 | pxor %xmm3, %xmm2\r | |
| 1498 | pxor %xmm3, %xmm4\r | |
| 1499 | pxor %xmm3, %xmm5\r | |
| 1500 | \r | |
| 1501 | movups 192(ctx), %xmm7\r | |
| 1502 | aesdec %xmm6, %xmm1\r | |
| 1503 | aesdec %xmm6, %xmm2\r | |
| 1504 | aesdec %xmm6, %xmm4\r | |
| 1505 | aesdec %xmm6, %xmm5\r | |
| 1506 | \r | |
| 1507 | movups 176(ctx), %xmm6\r | |
| 1508 | aesdec %xmm7, %xmm1\r | |
| 1509 | aesdec %xmm7, %xmm2\r | |
| 1510 | aesdec %xmm7, %xmm4\r | |
| 1511 | aesdec %xmm7, %xmm5\r | |
| 1512 | \r | |
| 1513 | movups 160(ctx), %xmm7\r | |
| 1514 | aesdec %xmm6, %xmm1\r | |
| 1515 | aesdec %xmm6, %xmm2\r | |
| 1516 | aesdec %xmm6, %xmm4\r | |
| 1517 | aesdec %xmm6, %xmm5\r | |
| 1518 | \r | |
| 1519 | movups 144(ctx), %xmm6\r | |
| 1520 | aesdec %xmm7, %xmm1\r | |
| 1521 | aesdec %xmm7, %xmm2\r | |
| 1522 | aesdec %xmm7, %xmm4\r | |
| 1523 | aesdec %xmm7, %xmm5\r | |
| 1524 | \r | |
| 1525 | movups 128(ctx), %xmm7\r | |
| 1526 | aesdec %xmm6, %xmm1\r | |
| 1527 | aesdec %xmm6, %xmm2\r | |
| 1528 | aesdec %xmm6, %xmm4\r | |
| 1529 | aesdec %xmm6, %xmm5\r | |
| 1530 | \r | |
| 1531 | movups 112(ctx), %xmm6\r | |
| 1532 | aesdec %xmm7, %xmm1\r | |
| 1533 | aesdec %xmm7, %xmm2\r | |
| 1534 | aesdec %xmm7, %xmm4\r | |
| 1535 | aesdec %xmm7, %xmm5\r | |
| 1536 | \r | |
| 1537 | movups 96(ctx), %xmm7\r | |
| 1538 | aesdec %xmm6, %xmm1\r | |
| 1539 | aesdec %xmm6, %xmm2\r | |
| 1540 | aesdec %xmm6, %xmm4\r | |
| 1541 | aesdec %xmm6, %xmm5\r | |
| 1542 | \r | |
| 1543 | movups 80(ctx), %xmm6\r | |
| 1544 | aesdec %xmm7, %xmm1\r | |
| 1545 | aesdec %xmm7, %xmm2\r | |
| 1546 | aesdec %xmm7, %xmm4\r | |
| 1547 | aesdec %xmm7, %xmm5\r | |
| 1548 | \r | |
| 1549 | movups 64(ctx), %xmm7\r | |
| 1550 | aesdec %xmm6, %xmm1\r | |
| 1551 | aesdec %xmm6, %xmm2\r | |
| 1552 | aesdec %xmm6, %xmm4\r | |
| 1553 | aesdec %xmm6, %xmm5\r | |
| 1554 | \r | |
| 1555 | movups 48(ctx), %xmm6\r | |
| 1556 | aesdec %xmm7, %xmm1\r | |
| 1557 | aesdec %xmm7, %xmm2\r | |
| 1558 | aesdec %xmm7, %xmm4\r | |
| 1559 | aesdec %xmm7, %xmm5\r | |
| 1560 | \r | |
| 1561 | movups 32(ctx), %xmm7\r | |
| 1562 | aesdec %xmm6, %xmm1\r | |
| 1563 | aesdec %xmm6, %xmm2\r | |
| 1564 | aesdec %xmm6, %xmm4\r | |
| 1565 | aesdec %xmm6, %xmm5\r | |
| 1566 | \r | |
| 1567 | movups 16(ctx), %xmm6\r | |
| 1568 | aesdec %xmm7, %xmm1\r | |
| 1569 | aesdec %xmm7, %xmm2\r | |
| 1570 | aesdec %xmm7, %xmm4\r | |
| 1571 | aesdec %xmm7, %xmm5\r | |
| 1572 | \r | |
| 1573 | movups 0(ctx), %xmm7\r | |
| 1574 | aesdec %xmm6, %xmm1\r | |
| 1575 | aesdec %xmm6, %xmm2\r | |
| 1576 | aesdec %xmm6, %xmm4\r | |
| 1577 | aesdec %xmm6, %xmm5\r | |
| 1578 | \r | |
| 1579 | aesdeclast %xmm7, %xmm1\r | |
| 1580 | aesdeclast %xmm7, %xmm2\r | |
| 1581 | aesdeclast %xmm7, %xmm4\r | |
| 1582 | aesdeclast %xmm7, %xmm5\r | |
| 1583 | \r | |
| 1584 | pxor iv, %xmm1 // 1st obuf ^= iv; \r | |
| 1585 | movups (ibuf), iv // 1st memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1586 | pxor iv, %xmm2 // 2nd obuf ^= iv; \r | |
| 1587 | movups 16(ibuf), iv // 2nd memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1588 | pxor iv, %xmm4 // 3rd obuf ^= iv; \r | |
| 1589 | movups 32(ibuf), iv // 3rd memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1590 | pxor iv, %xmm5 // 4th obuf ^= iv; \r | |
| 1591 | movups 48(ibuf), iv // 4th memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1592 | movups %xmm1, (obuf) // write 1st obuf\r | |
| 1593 | movups %xmm2, 16(obuf) // write 2nd obuf\r | |
| 1594 | movups %xmm4, 32(obuf) // write 3rd obuf\r | |
| 1595 | movups %xmm5, 48(obuf) // write 4th obuf\r | |
| 1596 | \r | |
| 1597 | add $64, ibuf // ibuf += AES_BLOCK_SIZE * 4; \r | |
| 1598 | add $64, obuf // obuf += AES_BLOCK_SIZE * 4; \r | |
| 1599 | \r | |
| 1600 | sub $4, num_blk // num_blk -= 4\r | |
| 1601 | jge 0b // if num_blk > 0, repeat the loop\r | |
| 1602 | \r | |
| 1603 | \r | |
| 1604 | 9: add $4, num_blk // post incremtn num_blk by 4\r | |
| 1605 | je L_HW_cbc_done // if num_blk == 0, no need for forthur processing code\r | |
| 1606 | \r | |
| 1607 | movups 208(ctx), %xmm4\r | |
| 1608 | movups 192(ctx), %xmm5\r | |
| 1609 | movups 176(ctx), %xmm6\r | |
| 1610 | movups 160(ctx), %xmm7\r | |
| 1611 | \r | |
| 1612 | #endif\r | |
| 1613 | \r | |
| 1614 | 0:\r | |
| 1615 | movups (ibuf), %xmm2 // tmp = ibuf\r | |
| 1616 | \r | |
| 1617 | // aes_decrypt\r | |
| 1618 | pxor %xmm3, %xmm2\r | |
| 1619 | aesdec %xmm4, %xmm2\r | |
| 1620 | aesdec %xmm5, %xmm2\r | |
| 1621 | aesdec %xmm6, %xmm2\r | |
| 1622 | aesdec %xmm7, %xmm2\r | |
| 1623 | #if defined __x86_64__\r | |
| 1624 | aesdec %xmm8, %xmm2\r | |
| 1625 | aesdec %xmm9, %xmm2\r | |
| 1626 | aesdec %xmm10, %xmm2\r | |
| 1627 | aesdec %xmm11, %xmm2\r | |
| 1628 | aesdec %xmm12, %xmm2\r | |
| 1629 | aesdec %xmm13, %xmm2\r | |
| 1630 | aesdec %xmm14, %xmm2\r | |
| 1631 | aesdec %xmm15, %xmm2\r | |
| 1632 | #else\r | |
| 1633 | movups 144(ctx), %xmm1\r | |
| 1634 | aesdec %xmm1, %xmm2\r | |
| 1635 | movups 128(ctx), %xmm1\r | |
| 1636 | aesdec %xmm1, %xmm2\r | |
| 1637 | movups 112(ctx), %xmm1\r | |
| 1638 | aesdec %xmm1, %xmm2\r | |
| 1639 | movups 96(ctx), %xmm1\r | |
| 1640 | aesdec %xmm1, %xmm2\r | |
| 1641 | movups 80(ctx), %xmm1\r | |
| 1642 | aesdec %xmm1, %xmm2\r | |
| 1643 | movups 64(ctx), %xmm1\r | |
| 1644 | aesdec %xmm1, %xmm2\r | |
| 1645 | movups 48(ctx), %xmm1\r | |
| 1646 | aesdec %xmm1, %xmm2\r | |
| 1647 | movups 32(ctx), %xmm1\r | |
| 1648 | aesdec %xmm1, %xmm2\r | |
| 1649 | #endif\r | |
| 1650 | movups 16(ctx), %xmm1\r | |
| 1651 | aesdec %xmm1, %xmm2\r | |
| 1652 | movups (ctx), %xmm1\r | |
| 1653 | aesdeclast %xmm1, %xmm2\r | |
| 1654 | \r | |
| 1655 | pxor iv, %xmm2 // obuf ^= iv; \r | |
| 1656 | movups (ibuf), iv // memcpy(iv, tmp, AES_BLOCK_SIZE);\r | |
| 1657 | \r | |
| 1658 | movups %xmm2, (obuf) // write obuf\r | |
| 1659 | \r | |
| 1660 | add $16, ibuf // ibuf += AES_BLOCK_SIZE; \r | |
| 1661 | add $16, obuf // obuf += AES_BLOCK_SIZE; \r | |
| 1662 | sub $1, num_blk // num_blk --\r | |
| 1663 | jg 0b // if num_blk > 0, repeat the loop\r | |
| 1664 | \r | |
| 1665 | jmp L_HW_cbc_done\r | |
| 1666 | \r | |
| 1667 | //\r | |
| 1668 | // --------- END of aes_decrypt_cbc_hw -------------------\r | |
| 1669 | //\r |