Mark Radbourne / mbedtls

mbed TLS Build

Dependents:   Encrypt_Decrypt1 mbed_blink_tls encrypt encrypt

library/aesni.c@0:cdf462088d13, 2017-01-05 (annotated)

Committer:
markrad
Date:
Thu Jan 05 00:18:44 2017 +0000
Revision:
0:cdf462088d13
Initial commit

Who changed what in which revision?

UserRevisionLine numberNew contents of line
markrad 0:cdf462088d13 1 /*
markrad 0:cdf462088d13 2 * AES-NI support functions
markrad 0:cdf462088d13 3 *
markrad 0:cdf462088d13 4 * Copyright (C) 2006-2015, ARM Limited, All Rights Reserved
markrad 0:cdf462088d13 5 * SPDX-License-Identifier: Apache-2.0
markrad 0:cdf462088d13 6 *
markrad 0:cdf462088d13 7 * Licensed under the Apache License, Version 2.0 (the "License"); you may
markrad 0:cdf462088d13 8 * not use this file except in compliance with the License.
markrad 0:cdf462088d13 9 * You may obtain a copy of the License at
markrad 0:cdf462088d13 10 *
markrad 0:cdf462088d13 11 * http://www.apache.org/licenses/LICENSE-2.0
markrad 0:cdf462088d13 12 *
markrad 0:cdf462088d13 13 * Unless required by applicable law or agreed to in writing, software
markrad 0:cdf462088d13 14 * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
markrad 0:cdf462088d13 15 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
markrad 0:cdf462088d13 16 * See the License for the specific language governing permissions and
markrad 0:cdf462088d13 17 * limitations under the License.
markrad 0:cdf462088d13 18 *
markrad 0:cdf462088d13 19 * This file is part of mbed TLS (https://tls.mbed.org)
markrad 0:cdf462088d13 20 */
markrad 0:cdf462088d13 21
markrad 0:cdf462088d13 22 /*
markrad 0:cdf462088d13 23 * [AES-WP] http://software.intel.com/en-us/articles/intel-advanced-encryption-standard-aes-instructions-set
markrad 0:cdf462088d13 24 * [CLMUL-WP] http://software.intel.com/en-us/articles/intel-carry-less-multiplication-instruction-and-its-usage-for-computing-the-gcm-mode/
markrad 0:cdf462088d13 25 */
markrad 0:cdf462088d13 26
markrad 0:cdf462088d13 27 #if !defined(MBEDTLS_CONFIG_FILE)
markrad 0:cdf462088d13 28 #include "mbedtls/config.h"
markrad 0:cdf462088d13 29 #else
markrad 0:cdf462088d13 30 #include MBEDTLS_CONFIG_FILE
markrad 0:cdf462088d13 31 #endif
markrad 0:cdf462088d13 32
markrad 0:cdf462088d13 33 #if defined(MBEDTLS_AESNI_C)
markrad 0:cdf462088d13 34
markrad 0:cdf462088d13 35 #include "mbedtls/aesni.h"
markrad 0:cdf462088d13 36
markrad 0:cdf462088d13 37 #include <string.h>
markrad 0:cdf462088d13 38
markrad 0:cdf462088d13 39 #ifndef asm
markrad 0:cdf462088d13 40 #define asm __asm
markrad 0:cdf462088d13 41 #endif
markrad 0:cdf462088d13 42
markrad 0:cdf462088d13 43 #if defined(MBEDTLS_HAVE_X86_64)
markrad 0:cdf462088d13 44
markrad 0:cdf462088d13 45 /*
markrad 0:cdf462088d13 46 * AES-NI support detection routine
markrad 0:cdf462088d13 47 */
markrad 0:cdf462088d13 48 int mbedtls_aesni_has_support( unsigned int what )
markrad 0:cdf462088d13 49 {
markrad 0:cdf462088d13 50 static int done = 0;
markrad 0:cdf462088d13 51 static unsigned int c = 0;
markrad 0:cdf462088d13 52
markrad 0:cdf462088d13 53 if( ! done )
markrad 0:cdf462088d13 54 {
markrad 0:cdf462088d13 55 asm( "movl $1, %%eax \n\t"
markrad 0:cdf462088d13 56 "cpuid \n\t"
markrad 0:cdf462088d13 57 : "=c" (c)
markrad 0:cdf462088d13 58 :
markrad 0:cdf462088d13 59 : "eax", "ebx", "edx" );
markrad 0:cdf462088d13 60 done = 1;
markrad 0:cdf462088d13 61 }
markrad 0:cdf462088d13 62
markrad 0:cdf462088d13 63 return( ( c & what ) != 0 );
markrad 0:cdf462088d13 64 }
markrad 0:cdf462088d13 65
markrad 0:cdf462088d13 66 /*
markrad 0:cdf462088d13 67 * Binutils needs to be at least 2.19 to support AES-NI instructions.
markrad 0:cdf462088d13 68 * Unfortunately, a lot of users have a lower version now (2014-04).
markrad 0:cdf462088d13 69 * Emit bytecode directly in order to support "old" version of gas.
markrad 0:cdf462088d13 70 *
markrad 0:cdf462088d13 71 * Opcodes from the Intel architecture reference manual, vol. 3.
markrad 0:cdf462088d13 72 * We always use registers, so we don't need prefixes for memory operands.
markrad 0:cdf462088d13 73 * Operand macros are in gas order (src, dst) as opposed to Intel order
markrad 0:cdf462088d13 74 * (dst, src) in order to blend better into the surrounding assembly code.
markrad 0:cdf462088d13 75 */
markrad 0:cdf462088d13 76 #define AESDEC ".byte 0x66,0x0F,0x38,0xDE,"
markrad 0:cdf462088d13 77 #define AESDECLAST ".byte 0x66,0x0F,0x38,0xDF,"
markrad 0:cdf462088d13 78 #define AESENC ".byte 0x66,0x0F,0x38,0xDC,"
markrad 0:cdf462088d13 79 #define AESENCLAST ".byte 0x66,0x0F,0x38,0xDD,"
markrad 0:cdf462088d13 80 #define AESIMC ".byte 0x66,0x0F,0x38,0xDB,"
markrad 0:cdf462088d13 81 #define AESKEYGENA ".byte 0x66,0x0F,0x3A,0xDF,"
markrad 0:cdf462088d13 82 #define PCLMULQDQ ".byte 0x66,0x0F,0x3A,0x44,"
markrad 0:cdf462088d13 83
markrad 0:cdf462088d13 84 #define xmm0_xmm0 "0xC0"
markrad 0:cdf462088d13 85 #define xmm0_xmm1 "0xC8"
markrad 0:cdf462088d13 86 #define xmm0_xmm2 "0xD0"
markrad 0:cdf462088d13 87 #define xmm0_xmm3 "0xD8"
markrad 0:cdf462088d13 88 #define xmm0_xmm4 "0xE0"
markrad 0:cdf462088d13 89 #define xmm1_xmm0 "0xC1"
markrad 0:cdf462088d13 90 #define xmm1_xmm2 "0xD1"
markrad 0:cdf462088d13 91
markrad 0:cdf462088d13 92 /*
markrad 0:cdf462088d13 93 * AES-NI AES-ECB block en(de)cryption
markrad 0:cdf462088d13 94 */
markrad 0:cdf462088d13 95 int mbedtls_aesni_crypt_ecb( mbedtls_aes_context *ctx,
markrad 0:cdf462088d13 96 int mode,
markrad 0:cdf462088d13 97 const unsigned char input[16],
markrad 0:cdf462088d13 98 unsigned char output[16] )
markrad 0:cdf462088d13 99 {
markrad 0:cdf462088d13 100 asm( "movdqu (%3), %%xmm0 \n\t" // load input
markrad 0:cdf462088d13 101 "movdqu (%1), %%xmm1 \n\t" // load round key 0
markrad 0:cdf462088d13 102 "pxor %%xmm1, %%xmm0 \n\t" // round 0
markrad 0:cdf462088d13 103 "add $16, %1 \n\t" // point to next round key
markrad 0:cdf462088d13 104 "subl $1, %0 \n\t" // normal rounds = nr - 1
markrad 0:cdf462088d13 105 "test %2, %2 \n\t" // mode?
markrad 0:cdf462088d13 106 "jz 2f \n\t" // 0 = decrypt
markrad 0:cdf462088d13 107
markrad 0:cdf462088d13 108 "1: \n\t" // encryption loop
markrad 0:cdf462088d13 109 "movdqu (%1), %%xmm1 \n\t" // load round key
markrad 0:cdf462088d13 110 AESENC xmm1_xmm0 "\n\t" // do round
markrad 0:cdf462088d13 111 "add $16, %1 \n\t" // point to next round key
markrad 0:cdf462088d13 112 "subl $1, %0 \n\t" // loop
markrad 0:cdf462088d13 113 "jnz 1b \n\t"
markrad 0:cdf462088d13 114 "movdqu (%1), %%xmm1 \n\t" // load round key
markrad 0:cdf462088d13 115 AESENCLAST xmm1_xmm0 "\n\t" // last round
markrad 0:cdf462088d13 116 "jmp 3f \n\t"
markrad 0:cdf462088d13 117
markrad 0:cdf462088d13 118 "2: \n\t" // decryption loop
markrad 0:cdf462088d13 119 "movdqu (%1), %%xmm1 \n\t"
markrad 0:cdf462088d13 120 AESDEC xmm1_xmm0 "\n\t" // do round
markrad 0:cdf462088d13 121 "add $16, %1 \n\t"
markrad 0:cdf462088d13 122 "subl $1, %0 \n\t"
markrad 0:cdf462088d13 123 "jnz 2b \n\t"
markrad 0:cdf462088d13 124 "movdqu (%1), %%xmm1 \n\t" // load round key
markrad 0:cdf462088d13 125 AESDECLAST xmm1_xmm0 "\n\t" // last round
markrad 0:cdf462088d13 126
markrad 0:cdf462088d13 127 "3: \n\t"
markrad 0:cdf462088d13 128 "movdqu %%xmm0, (%4) \n\t" // export output
markrad 0:cdf462088d13 129 :
markrad 0:cdf462088d13 130 : "r" (ctx->nr), "r" (ctx->rk), "r" (mode), "r" (input), "r" (output)
markrad 0:cdf462088d13 131 : "memory", "cc", "xmm0", "xmm1" );
markrad 0:cdf462088d13 132
markrad 0:cdf462088d13 133
markrad 0:cdf462088d13 134 return( 0 );
markrad 0:cdf462088d13 135 }
markrad 0:cdf462088d13 136
markrad 0:cdf462088d13 137 /*
markrad 0:cdf462088d13 138 * GCM multiplication: c = a times b in GF(2^128)
markrad 0:cdf462088d13 139 * Based on [CLMUL-WP] algorithms 1 (with equation 27) and 5.
markrad 0:cdf462088d13 140 */
markrad 0:cdf462088d13 141 void mbedtls_aesni_gcm_mult( unsigned char c[16],
markrad 0:cdf462088d13 142 const unsigned char a[16],
markrad 0:cdf462088d13 143 const unsigned char b[16] )
markrad 0:cdf462088d13 144 {
markrad 0:cdf462088d13 145 unsigned char aa[16], bb[16], cc[16];
markrad 0:cdf462088d13 146 size_t i;
markrad 0:cdf462088d13 147
markrad 0:cdf462088d13 148 /* The inputs are in big-endian order, so byte-reverse them */
markrad 0:cdf462088d13 149 for( i = 0; i < 16; i++ )
markrad 0:cdf462088d13 150 {
markrad 0:cdf462088d13 151 aa[i] = a[15 - i];
markrad 0:cdf462088d13 152 bb[i] = b[15 - i];
markrad 0:cdf462088d13 153 }
markrad 0:cdf462088d13 154
markrad 0:cdf462088d13 155 asm( "movdqu (%0), %%xmm0 \n\t" // a1:a0
markrad 0:cdf462088d13 156 "movdqu (%1), %%xmm1 \n\t" // b1:b0
markrad 0:cdf462088d13 157
markrad 0:cdf462088d13 158 /*
markrad 0:cdf462088d13 159 * Caryless multiplication xmm2:xmm1 = xmm0 * xmm1
markrad 0:cdf462088d13 160 * using [CLMUL-WP] algorithm 1 (p. 13).
markrad 0:cdf462088d13 161 */
markrad 0:cdf462088d13 162 "movdqa %%xmm1, %%xmm2 \n\t" // copy of b1:b0
markrad 0:cdf462088d13 163 "movdqa %%xmm1, %%xmm3 \n\t" // same
markrad 0:cdf462088d13 164 "movdqa %%xmm1, %%xmm4 \n\t" // same
markrad 0:cdf462088d13 165 PCLMULQDQ xmm0_xmm1 ",0x00 \n\t" // a0*b0 = c1:c0
markrad 0:cdf462088d13 166 PCLMULQDQ xmm0_xmm2 ",0x11 \n\t" // a1*b1 = d1:d0
markrad 0:cdf462088d13 167 PCLMULQDQ xmm0_xmm3 ",0x10 \n\t" // a0*b1 = e1:e0
markrad 0:cdf462088d13 168 PCLMULQDQ xmm0_xmm4 ",0x01 \n\t" // a1*b0 = f1:f0
markrad 0:cdf462088d13 169 "pxor %%xmm3, %%xmm4 \n\t" // e1+f1:e0+f0
markrad 0:cdf462088d13 170 "movdqa %%xmm4, %%xmm3 \n\t" // same
markrad 0:cdf462088d13 171 "psrldq $8, %%xmm4 \n\t" // 0:e1+f1
markrad 0:cdf462088d13 172 "pslldq $8, %%xmm3 \n\t" // e0+f0:0
markrad 0:cdf462088d13 173 "pxor %%xmm4, %%xmm2 \n\t" // d1:d0+e1+f1
markrad 0:cdf462088d13 174 "pxor %%xmm3, %%xmm1 \n\t" // c1+e0+f1:c0
markrad 0:cdf462088d13 175
markrad 0:cdf462088d13 176 /*
markrad 0:cdf462088d13 177 * Now shift the result one bit to the left,
markrad 0:cdf462088d13 178 * taking advantage of [CLMUL-WP] eq 27 (p. 20)
markrad 0:cdf462088d13 179 */
markrad 0:cdf462088d13 180 "movdqa %%xmm1, %%xmm3 \n\t" // r1:r0
markrad 0:cdf462088d13 181 "movdqa %%xmm2, %%xmm4 \n\t" // r3:r2
markrad 0:cdf462088d13 182 "psllq $1, %%xmm1 \n\t" // r1<<1:r0<<1
markrad 0:cdf462088d13 183 "psllq $1, %%xmm2 \n\t" // r3<<1:r2<<1
markrad 0:cdf462088d13 184 "psrlq $63, %%xmm3 \n\t" // r1>>63:r0>>63
markrad 0:cdf462088d13 185 "psrlq $63, %%xmm4 \n\t" // r3>>63:r2>>63
markrad 0:cdf462088d13 186 "movdqa %%xmm3, %%xmm5 \n\t" // r1>>63:r0>>63
markrad 0:cdf462088d13 187 "pslldq $8, %%xmm3 \n\t" // r0>>63:0
markrad 0:cdf462088d13 188 "pslldq $8, %%xmm4 \n\t" // r2>>63:0
markrad 0:cdf462088d13 189 "psrldq $8, %%xmm5 \n\t" // 0:r1>>63
markrad 0:cdf462088d13 190 "por %%xmm3, %%xmm1 \n\t" // r1<<1|r0>>63:r0<<1
markrad 0:cdf462088d13 191 "por %%xmm4, %%xmm2 \n\t" // r3<<1|r2>>62:r2<<1
markrad 0:cdf462088d13 192 "por %%xmm5, %%xmm2 \n\t" // r3<<1|r2>>62:r2<<1|r1>>63
markrad 0:cdf462088d13 193
markrad 0:cdf462088d13 194 /*
markrad 0:cdf462088d13 195 * Now reduce modulo the GCM polynomial x^128 + x^7 + x^2 + x + 1
markrad 0:cdf462088d13 196 * using [CLMUL-WP] algorithm 5 (p. 20).
markrad 0:cdf462088d13 197 * Currently xmm2:xmm1 holds x3:x2:x1:x0 (already shifted).
markrad 0:cdf462088d13 198 */
markrad 0:cdf462088d13 199 /* Step 2 (1) */
markrad 0:cdf462088d13 200 "movdqa %%xmm1, %%xmm3 \n\t" // x1:x0
markrad 0:cdf462088d13 201 "movdqa %%xmm1, %%xmm4 \n\t" // same
markrad 0:cdf462088d13 202 "movdqa %%xmm1, %%xmm5 \n\t" // same
markrad 0:cdf462088d13 203 "psllq $63, %%xmm3 \n\t" // x1<<63:x0<<63 = stuff:a
markrad 0:cdf462088d13 204 "psllq $62, %%xmm4 \n\t" // x1<<62:x0<<62 = stuff:b
markrad 0:cdf462088d13 205 "psllq $57, %%xmm5 \n\t" // x1<<57:x0<<57 = stuff:c
markrad 0:cdf462088d13 206
markrad 0:cdf462088d13 207 /* Step 2 (2) */
markrad 0:cdf462088d13 208 "pxor %%xmm4, %%xmm3 \n\t" // stuff:a+b
markrad 0:cdf462088d13 209 "pxor %%xmm5, %%xmm3 \n\t" // stuff:a+b+c
markrad 0:cdf462088d13 210 "pslldq $8, %%xmm3 \n\t" // a+b+c:0
markrad 0:cdf462088d13 211 "pxor %%xmm3, %%xmm1 \n\t" // x1+a+b+c:x0 = d:x0
markrad 0:cdf462088d13 212
markrad 0:cdf462088d13 213 /* Steps 3 and 4 */
markrad 0:cdf462088d13 214 "movdqa %%xmm1,%%xmm0 \n\t" // d:x0
markrad 0:cdf462088d13 215 "movdqa %%xmm1,%%xmm4 \n\t" // same
markrad 0:cdf462088d13 216 "movdqa %%xmm1,%%xmm5 \n\t" // same
markrad 0:cdf462088d13 217 "psrlq $1, %%xmm0 \n\t" // e1:x0>>1 = e1:e0'
markrad 0:cdf462088d13 218 "psrlq $2, %%xmm4 \n\t" // f1:x0>>2 = f1:f0'
markrad 0:cdf462088d13 219 "psrlq $7, %%xmm5 \n\t" // g1:x0>>7 = g1:g0'
markrad 0:cdf462088d13 220 "pxor %%xmm4, %%xmm0 \n\t" // e1+f1:e0'+f0'
markrad 0:cdf462088d13 221 "pxor %%xmm5, %%xmm0 \n\t" // e1+f1+g1:e0'+f0'+g0'
markrad 0:cdf462088d13 222 // e0'+f0'+g0' is almost e0+f0+g0, ex\tcept for some missing
markrad 0:cdf462088d13 223 // bits carried from d. Now get those\t bits back in.
markrad 0:cdf462088d13 224 "movdqa %%xmm1,%%xmm3 \n\t" // d:x0
markrad 0:cdf462088d13 225 "movdqa %%xmm1,%%xmm4 \n\t" // same
markrad 0:cdf462088d13 226 "movdqa %%xmm1,%%xmm5 \n\t" // same
markrad 0:cdf462088d13 227 "psllq $63, %%xmm3 \n\t" // d<<63:stuff
markrad 0:cdf462088d13 228 "psllq $62, %%xmm4 \n\t" // d<<62:stuff
markrad 0:cdf462088d13 229 "psllq $57, %%xmm5 \n\t" // d<<57:stuff
markrad 0:cdf462088d13 230 "pxor %%xmm4, %%xmm3 \n\t" // d<<63+d<<62:stuff
markrad 0:cdf462088d13 231 "pxor %%xmm5, %%xmm3 \n\t" // missing bits of d:stuff
markrad 0:cdf462088d13 232 "psrldq $8, %%xmm3 \n\t" // 0:missing bits of d
markrad 0:cdf462088d13 233 "pxor %%xmm3, %%xmm0 \n\t" // e1+f1+g1:e0+f0+g0
markrad 0:cdf462088d13 234 "pxor %%xmm1, %%xmm0 \n\t" // h1:h0
markrad 0:cdf462088d13 235 "pxor %%xmm2, %%xmm0 \n\t" // x3+h1:x2+h0
markrad 0:cdf462088d13 236
markrad 0:cdf462088d13 237 "movdqu %%xmm0, (%2) \n\t" // done
markrad 0:cdf462088d13 238 :
markrad 0:cdf462088d13 239 : "r" (aa), "r" (bb), "r" (cc)
markrad 0:cdf462088d13 240 : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" );
markrad 0:cdf462088d13 241
markrad 0:cdf462088d13 242 /* Now byte-reverse the outputs */
markrad 0:cdf462088d13 243 for( i = 0; i < 16; i++ )
markrad 0:cdf462088d13 244 c[i] = cc[15 - i];
markrad 0:cdf462088d13 245
markrad 0:cdf462088d13 246 return;
markrad 0:cdf462088d13 247 }
markrad 0:cdf462088d13 248
markrad 0:cdf462088d13 249 /*
markrad 0:cdf462088d13 250 * Compute decryption round keys from encryption round keys
markrad 0:cdf462088d13 251 */
markrad 0:cdf462088d13 252 void mbedtls_aesni_inverse_key( unsigned char *invkey,
markrad 0:cdf462088d13 253 const unsigned char *fwdkey, int nr )
markrad 0:cdf462088d13 254 {
markrad 0:cdf462088d13 255 unsigned char *ik = invkey;
markrad 0:cdf462088d13 256 const unsigned char *fk = fwdkey + 16 * nr;
markrad 0:cdf462088d13 257
markrad 0:cdf462088d13 258 memcpy( ik, fk, 16 );
markrad 0:cdf462088d13 259
markrad 0:cdf462088d13 260 for( fk -= 16, ik += 16; fk > fwdkey; fk -= 16, ik += 16 )
markrad 0:cdf462088d13 261 asm( "movdqu (%0), %%xmm0 \n\t"
markrad 0:cdf462088d13 262 AESIMC xmm0_xmm0 "\n\t"
markrad 0:cdf462088d13 263 "movdqu %%xmm0, (%1) \n\t"
markrad 0:cdf462088d13 264 :
markrad 0:cdf462088d13 265 : "r" (fk), "r" (ik)
markrad 0:cdf462088d13 266 : "memory", "xmm0" );
markrad 0:cdf462088d13 267
markrad 0:cdf462088d13 268 memcpy( ik, fk, 16 );
markrad 0:cdf462088d13 269 }
markrad 0:cdf462088d13 270
markrad 0:cdf462088d13 271 /*
markrad 0:cdf462088d13 272 * Key expansion, 128-bit case
markrad 0:cdf462088d13 273 */
markrad 0:cdf462088d13 274 static void aesni_setkey_enc_128( unsigned char *rk,
markrad 0:cdf462088d13 275 const unsigned char *key )
markrad 0:cdf462088d13 276 {
markrad 0:cdf462088d13 277 asm( "movdqu (%1), %%xmm0 \n\t" // copy the original key
markrad 0:cdf462088d13 278 "movdqu %%xmm0, (%0) \n\t" // as round key 0
markrad 0:cdf462088d13 279 "jmp 2f \n\t" // skip auxiliary routine
markrad 0:cdf462088d13 280
markrad 0:cdf462088d13 281 /*
markrad 0:cdf462088d13 282 * Finish generating the next round key.
markrad 0:cdf462088d13 283 *
markrad 0:cdf462088d13 284 * On entry xmm0 is r3:r2:r1:r0 and xmm1 is X:stuff:stuff:stuff
markrad 0:cdf462088d13 285 * with X = rot( sub( r3 ) ) ^ RCON.
markrad 0:cdf462088d13 286 *
markrad 0:cdf462088d13 287 * On exit, xmm0 is r7:r6:r5:r4
markrad 0:cdf462088d13 288 * with r4 = X + r0, r5 = r4 + r1, r6 = r5 + r2, r7 = r6 + r3
markrad 0:cdf462088d13 289 * and those are written to the round key buffer.
markrad 0:cdf462088d13 290 */
markrad 0:cdf462088d13 291 "1: \n\t"
markrad 0:cdf462088d13 292 "pshufd $0xff, %%xmm1, %%xmm1 \n\t" // X:X:X:X
markrad 0:cdf462088d13 293 "pxor %%xmm0, %%xmm1 \n\t" // X+r3:X+r2:X+r1:r4
markrad 0:cdf462088d13 294 "pslldq $4, %%xmm0 \n\t" // r2:r1:r0:0
markrad 0:cdf462088d13 295 "pxor %%xmm0, %%xmm1 \n\t" // X+r3+r2:X+r2+r1:r5:r4
markrad 0:cdf462088d13 296 "pslldq $4, %%xmm0 \n\t" // etc
markrad 0:cdf462088d13 297 "pxor %%xmm0, %%xmm1 \n\t"
markrad 0:cdf462088d13 298 "pslldq $4, %%xmm0 \n\t"
markrad 0:cdf462088d13 299 "pxor %%xmm1, %%xmm0 \n\t" // update xmm0 for next time!
markrad 0:cdf462088d13 300 "add $16, %0 \n\t" // point to next round key
markrad 0:cdf462088d13 301 "movdqu %%xmm0, (%0) \n\t" // write it
markrad 0:cdf462088d13 302 "ret \n\t"
markrad 0:cdf462088d13 303
markrad 0:cdf462088d13 304 /* Main "loop" */
markrad 0:cdf462088d13 305 "2: \n\t"
markrad 0:cdf462088d13 306 AESKEYGENA xmm0_xmm1 ",0x01 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 307 AESKEYGENA xmm0_xmm1 ",0x02 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 308 AESKEYGENA xmm0_xmm1 ",0x04 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 309 AESKEYGENA xmm0_xmm1 ",0x08 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 310 AESKEYGENA xmm0_xmm1 ",0x10 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 311 AESKEYGENA xmm0_xmm1 ",0x20 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 312 AESKEYGENA xmm0_xmm1 ",0x40 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 313 AESKEYGENA xmm0_xmm1 ",0x80 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 314 AESKEYGENA xmm0_xmm1 ",0x1B \n\tcall 1b \n\t"
markrad 0:cdf462088d13 315 AESKEYGENA xmm0_xmm1 ",0x36 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 316 :
markrad 0:cdf462088d13 317 : "r" (rk), "r" (key)
markrad 0:cdf462088d13 318 : "memory", "cc", "0" );
markrad 0:cdf462088d13 319 }
markrad 0:cdf462088d13 320
markrad 0:cdf462088d13 321 /*
markrad 0:cdf462088d13 322 * Key expansion, 192-bit case
markrad 0:cdf462088d13 323 */
markrad 0:cdf462088d13 324 static void aesni_setkey_enc_192( unsigned char *rk,
markrad 0:cdf462088d13 325 const unsigned char *key )
markrad 0:cdf462088d13 326 {
markrad 0:cdf462088d13 327 asm( "movdqu (%1), %%xmm0 \n\t" // copy original round key
markrad 0:cdf462088d13 328 "movdqu %%xmm0, (%0) \n\t"
markrad 0:cdf462088d13 329 "add $16, %0 \n\t"
markrad 0:cdf462088d13 330 "movq 16(%1), %%xmm1 \n\t"
markrad 0:cdf462088d13 331 "movq %%xmm1, (%0) \n\t"
markrad 0:cdf462088d13 332 "add $8, %0 \n\t"
markrad 0:cdf462088d13 333 "jmp 2f \n\t" // skip auxiliary routine
markrad 0:cdf462088d13 334
markrad 0:cdf462088d13 335 /*
markrad 0:cdf462088d13 336 * Finish generating the next 6 quarter-keys.
markrad 0:cdf462088d13 337 *
markrad 0:cdf462088d13 338 * On entry xmm0 is r3:r2:r1:r0, xmm1 is stuff:stuff:r5:r4
markrad 0:cdf462088d13 339 * and xmm2 is stuff:stuff:X:stuff with X = rot( sub( r3 ) ) ^ RCON.
markrad 0:cdf462088d13 340 *
markrad 0:cdf462088d13 341 * On exit, xmm0 is r9:r8:r7:r6 and xmm1 is stuff:stuff:r11:r10
markrad 0:cdf462088d13 342 * and those are written to the round key buffer.
markrad 0:cdf462088d13 343 */
markrad 0:cdf462088d13 344 "1: \n\t"
markrad 0:cdf462088d13 345 "pshufd $0x55, %%xmm2, %%xmm2 \n\t" // X:X:X:X
markrad 0:cdf462088d13 346 "pxor %%xmm0, %%xmm2 \n\t" // X+r3:X+r2:X+r1:r4
markrad 0:cdf462088d13 347 "pslldq $4, %%xmm0 \n\t" // etc
markrad 0:cdf462088d13 348 "pxor %%xmm0, %%xmm2 \n\t"
markrad 0:cdf462088d13 349 "pslldq $4, %%xmm0 \n\t"
markrad 0:cdf462088d13 350 "pxor %%xmm0, %%xmm2 \n\t"
markrad 0:cdf462088d13 351 "pslldq $4, %%xmm0 \n\t"
markrad 0:cdf462088d13 352 "pxor %%xmm2, %%xmm0 \n\t" // update xmm0 = r9:r8:r7:r6
markrad 0:cdf462088d13 353 "movdqu %%xmm0, (%0) \n\t"
markrad 0:cdf462088d13 354 "add $16, %0 \n\t"
markrad 0:cdf462088d13 355 "pshufd $0xff, %%xmm0, %%xmm2 \n\t" // r9:r9:r9:r9
markrad 0:cdf462088d13 356 "pxor %%xmm1, %%xmm2 \n\t" // stuff:stuff:r9+r5:r10
markrad 0:cdf462088d13 357 "pslldq $4, %%xmm1 \n\t" // r2:r1:r0:0
markrad 0:cdf462088d13 358 "pxor %%xmm2, %%xmm1 \n\t" // xmm1 = stuff:stuff:r11:r10
markrad 0:cdf462088d13 359 "movq %%xmm1, (%0) \n\t"
markrad 0:cdf462088d13 360 "add $8, %0 \n\t"
markrad 0:cdf462088d13 361 "ret \n\t"
markrad 0:cdf462088d13 362
markrad 0:cdf462088d13 363 "2: \n\t"
markrad 0:cdf462088d13 364 AESKEYGENA xmm1_xmm2 ",0x01 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 365 AESKEYGENA xmm1_xmm2 ",0x02 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 366 AESKEYGENA xmm1_xmm2 ",0x04 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 367 AESKEYGENA xmm1_xmm2 ",0x08 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 368 AESKEYGENA xmm1_xmm2 ",0x10 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 369 AESKEYGENA xmm1_xmm2 ",0x20 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 370 AESKEYGENA xmm1_xmm2 ",0x40 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 371 AESKEYGENA xmm1_xmm2 ",0x80 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 372
markrad 0:cdf462088d13 373 :
markrad 0:cdf462088d13 374 : "r" (rk), "r" (key)
markrad 0:cdf462088d13 375 : "memory", "cc", "0" );
markrad 0:cdf462088d13 376 }
markrad 0:cdf462088d13 377
markrad 0:cdf462088d13 378 /*
markrad 0:cdf462088d13 379 * Key expansion, 256-bit case
markrad 0:cdf462088d13 380 */
markrad 0:cdf462088d13 381 static void aesni_setkey_enc_256( unsigned char *rk,
markrad 0:cdf462088d13 382 const unsigned char *key )
markrad 0:cdf462088d13 383 {
markrad 0:cdf462088d13 384 asm( "movdqu (%1), %%xmm0 \n\t"
markrad 0:cdf462088d13 385 "movdqu %%xmm0, (%0) \n\t"
markrad 0:cdf462088d13 386 "add $16, %0 \n\t"
markrad 0:cdf462088d13 387 "movdqu 16(%1), %%xmm1 \n\t"
markrad 0:cdf462088d13 388 "movdqu %%xmm1, (%0) \n\t"
markrad 0:cdf462088d13 389 "jmp 2f \n\t" // skip auxiliary routine
markrad 0:cdf462088d13 390
markrad 0:cdf462088d13 391 /*
markrad 0:cdf462088d13 392 * Finish generating the next two round keys.
markrad 0:cdf462088d13 393 *
markrad 0:cdf462088d13 394 * On entry xmm0 is r3:r2:r1:r0, xmm1 is r7:r6:r5:r4 and
markrad 0:cdf462088d13 395 * xmm2 is X:stuff:stuff:stuff with X = rot( sub( r7 )) ^ RCON
markrad 0:cdf462088d13 396 *
markrad 0:cdf462088d13 397 * On exit, xmm0 is r11:r10:r9:r8 and xmm1 is r15:r14:r13:r12
markrad 0:cdf462088d13 398 * and those have been written to the output buffer.
markrad 0:cdf462088d13 399 */
markrad 0:cdf462088d13 400 "1: \n\t"
markrad 0:cdf462088d13 401 "pshufd $0xff, %%xmm2, %%xmm2 \n\t"
markrad 0:cdf462088d13 402 "pxor %%xmm0, %%xmm2 \n\t"
markrad 0:cdf462088d13 403 "pslldq $4, %%xmm0 \n\t"
markrad 0:cdf462088d13 404 "pxor %%xmm0, %%xmm2 \n\t"
markrad 0:cdf462088d13 405 "pslldq $4, %%xmm0 \n\t"
markrad 0:cdf462088d13 406 "pxor %%xmm0, %%xmm2 \n\t"
markrad 0:cdf462088d13 407 "pslldq $4, %%xmm0 \n\t"
markrad 0:cdf462088d13 408 "pxor %%xmm2, %%xmm0 \n\t"
markrad 0:cdf462088d13 409 "add $16, %0 \n\t"
markrad 0:cdf462088d13 410 "movdqu %%xmm0, (%0) \n\t"
markrad 0:cdf462088d13 411
markrad 0:cdf462088d13 412 /* Set xmm2 to stuff:Y:stuff:stuff with Y = subword( r11 )
markrad 0:cdf462088d13 413 * and proceed to generate next round key from there */
markrad 0:cdf462088d13 414 AESKEYGENA xmm0_xmm2 ",0x00 \n\t"
markrad 0:cdf462088d13 415 "pshufd $0xaa, %%xmm2, %%xmm2 \n\t"
markrad 0:cdf462088d13 416 "pxor %%xmm1, %%xmm2 \n\t"
markrad 0:cdf462088d13 417 "pslldq $4, %%xmm1 \n\t"
markrad 0:cdf462088d13 418 "pxor %%xmm1, %%xmm2 \n\t"
markrad 0:cdf462088d13 419 "pslldq $4, %%xmm1 \n\t"
markrad 0:cdf462088d13 420 "pxor %%xmm1, %%xmm2 \n\t"
markrad 0:cdf462088d13 421 "pslldq $4, %%xmm1 \n\t"
markrad 0:cdf462088d13 422 "pxor %%xmm2, %%xmm1 \n\t"
markrad 0:cdf462088d13 423 "add $16, %0 \n\t"
markrad 0:cdf462088d13 424 "movdqu %%xmm1, (%0) \n\t"
markrad 0:cdf462088d13 425 "ret \n\t"
markrad 0:cdf462088d13 426
markrad 0:cdf462088d13 427 /*
markrad 0:cdf462088d13 428 * Main "loop" - Generating one more key than necessary,
markrad 0:cdf462088d13 429 * see definition of mbedtls_aes_context.buf
markrad 0:cdf462088d13 430 */
markrad 0:cdf462088d13 431 "2: \n\t"
markrad 0:cdf462088d13 432 AESKEYGENA xmm1_xmm2 ",0x01 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 433 AESKEYGENA xmm1_xmm2 ",0x02 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 434 AESKEYGENA xmm1_xmm2 ",0x04 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 435 AESKEYGENA xmm1_xmm2 ",0x08 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 436 AESKEYGENA xmm1_xmm2 ",0x10 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 437 AESKEYGENA xmm1_xmm2 ",0x20 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 438 AESKEYGENA xmm1_xmm2 ",0x40 \n\tcall 1b \n\t"
markrad 0:cdf462088d13 439 :
markrad 0:cdf462088d13 440 : "r" (rk), "r" (key)
markrad 0:cdf462088d13 441 : "memory", "cc", "0" );
markrad 0:cdf462088d13 442 }
markrad 0:cdf462088d13 443
markrad 0:cdf462088d13 444 /*
markrad 0:cdf462088d13 445 * Key expansion, wrapper
markrad 0:cdf462088d13 446 */
markrad 0:cdf462088d13 447 int mbedtls_aesni_setkey_enc( unsigned char *rk,
markrad 0:cdf462088d13 448 const unsigned char *key,
markrad 0:cdf462088d13 449 size_t bits )
markrad 0:cdf462088d13 450 {
markrad 0:cdf462088d13 451 switch( bits )
markrad 0:cdf462088d13 452 {
markrad 0:cdf462088d13 453 case 128: aesni_setkey_enc_128( rk, key ); break;
markrad 0:cdf462088d13 454 case 192: aesni_setkey_enc_192( rk, key ); break;
markrad 0:cdf462088d13 455 case 256: aesni_setkey_enc_256( rk, key ); break;
markrad 0:cdf462088d13 456 default : return( MBEDTLS_ERR_AES_INVALID_KEY_LENGTH );
markrad 0:cdf462088d13 457 }
markrad 0:cdf462088d13 458
markrad 0:cdf462088d13 459 return( 0 );
markrad 0:cdf462088d13 460 }
markrad 0:cdf462088d13 461
markrad 0:cdf462088d13 462 #endif /* MBEDTLS_HAVE_X86_64 */
markrad 0:cdf462088d13 463
markrad 0:cdf462088d13 464 #endif /* MBEDTLS_AESNI_C */