common/mbedtls/aesni.c

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