2 * AES-NI support functions
4 * Copyright (C) 2006-2015, ARM Limited, All Rights Reserved
5 * SPDX-License-Identifier: Apache-2.0
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
11 * http://www.apache.org/licenses/LICENSE-2.0
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.
19 * This file is part of mbed TLS (https://tls.mbed.org)
23 * [AES-WP] http://software.intel.com/en-us/articles/intel-advanced-encryption-standard-aes-instructions-set
24 * [CLMUL-WP] http://software.intel.com/en-us/articles/intel-carry-less-multiplication-instruction-and-its-usage-for-computing-the-gcm-mode/
27 #if !defined(MBEDTLS_CONFIG_FILE)
28 #include "mbedtls/config.h"
30 #include MBEDTLS_CONFIG_FILE
33 #if defined(MBEDTLS_AESNI_C)
35 #include "mbedtls/aesni.h"
43 #if defined(MBEDTLS_HAVE_X86_64)
46 * AES-NI support detection routine
48 int mbedtls_aesni_has_support( unsigned int what
)
51 static unsigned int c
= 0;
55 asm( "movl $1, %%eax \n\t"
59 : "eax", "ebx", "edx" );
63 return( ( c
& what
) != 0 );
67 * Binutils needs to be at least 2.19 to support AES-NI instructions.
68 * Unfortunately, a lot of users have a lower version now (2014-04).
69 * Emit bytecode directly in order to support "old" version of gas.
71 * Opcodes from the Intel architecture reference manual, vol. 3.
72 * We always use registers, so we don't need prefixes for memory operands.
73 * Operand macros are in gas order (src, dst) as opposed to Intel order
74 * (dst, src) in order to blend better into the surrounding assembly code.
76 #define AESDEC ".byte 0x66,0x0F,0x38,0xDE,"
77 #define AESDECLAST ".byte 0x66,0x0F,0x38,0xDF,"
78 #define AESENC ".byte 0x66,0x0F,0x38,0xDC,"
79 #define AESENCLAST ".byte 0x66,0x0F,0x38,0xDD,"
80 #define AESIMC ".byte 0x66,0x0F,0x38,0xDB,"
81 #define AESKEYGENA ".byte 0x66,0x0F,0x3A,0xDF,"
82 #define PCLMULQDQ ".byte 0x66,0x0F,0x3A,0x44,"
84 #define xmm0_xmm0 "0xC0"
85 #define xmm0_xmm1 "0xC8"
86 #define xmm0_xmm2 "0xD0"
87 #define xmm0_xmm3 "0xD8"
88 #define xmm0_xmm4 "0xE0"
89 #define xmm1_xmm0 "0xC1"
90 #define xmm1_xmm2 "0xD1"
93 * AES-NI AES-ECB block en(de)cryption
95 int mbedtls_aesni_crypt_ecb( mbedtls_aes_context
*ctx
,
97 const unsigned char input
[16],
98 unsigned char output
[16] )
100 asm( "movdqu (%3), %%xmm0 \n\t" // load input
101 "movdqu (%1), %%xmm1 \n\t" // load round key 0
102 "pxor %%xmm1, %%xmm0 \n\t" // round 0
103 "addq $16, %1 \n\t" // point to next round key
104 "subl $1, %0 \n\t" // normal rounds = nr - 1
105 "test %2, %2 \n\t" // mode?
106 "jz 2f \n\t" // 0 = decrypt
108 "1: \n\t" // encryption loop
109 "movdqu (%1), %%xmm1 \n\t" // load round key
110 AESENC xmm1_xmm0
"\n\t" // do round
111 "addq $16, %1 \n\t" // point to next round key
112 "subl $1, %0 \n\t" // loop
114 "movdqu (%1), %%xmm1 \n\t" // load round key
115 AESENCLAST xmm1_xmm0
"\n\t" // last round
118 "2: \n\t" // decryption loop
119 "movdqu (%1), %%xmm1 \n\t"
120 AESDEC xmm1_xmm0
"\n\t" // do round
124 "movdqu (%1), %%xmm1 \n\t" // load round key
125 AESDECLAST xmm1_xmm0
"\n\t" // last round
128 "movdqu %%xmm0, (%4) \n\t" // export output
130 : "r" (ctx
->nr
), "r" (ctx
->rk
), "r" (mode
), "r" (input
), "r" (output
)
131 : "memory", "cc", "xmm0", "xmm1" );
138 * GCM multiplication: c = a times b in GF(2^128)
139 * Based on [CLMUL-WP] algorithms 1 (with equation 27) and 5.
141 void mbedtls_aesni_gcm_mult( unsigned char c
[16],
142 const unsigned char a
[16],
143 const unsigned char b
[16] )
145 unsigned char aa
[16], bb
[16], cc
[16];
148 /* The inputs are in big-endian order, so byte-reverse them */
149 for( i
= 0; i
< 16; i
++ )
155 asm( "movdqu (%0), %%xmm0 \n\t" // a1:a0
156 "movdqu (%1), %%xmm1 \n\t" // b1:b0
159 * Caryless multiplication xmm2:xmm1 = xmm0 * xmm1
160 * using [CLMUL-WP] algorithm 1 (p. 13).
162 "movdqa %%xmm1, %%xmm2 \n\t" // copy of b1:b0
163 "movdqa %%xmm1, %%xmm3 \n\t" // same
164 "movdqa %%xmm1, %%xmm4 \n\t" // same
165 PCLMULQDQ xmm0_xmm1
",0x00 \n\t" // a0*b0 = c1:c0
166 PCLMULQDQ xmm0_xmm2
",0x11 \n\t" // a1*b1 = d1:d0
167 PCLMULQDQ xmm0_xmm3
",0x10 \n\t" // a0*b1 = e1:e0
168 PCLMULQDQ xmm0_xmm4
",0x01 \n\t" // a1*b0 = f1:f0
169 "pxor %%xmm3, %%xmm4 \n\t" // e1+f1:e0+f0
170 "movdqa %%xmm4, %%xmm3 \n\t" // same
171 "psrldq $8, %%xmm4 \n\t" // 0:e1+f1
172 "pslldq $8, %%xmm3 \n\t" // e0+f0:0
173 "pxor %%xmm4, %%xmm2 \n\t" // d1:d0+e1+f1
174 "pxor %%xmm3, %%xmm1 \n\t" // c1+e0+f1:c0
177 * Now shift the result one bit to the left,
178 * taking advantage of [CLMUL-WP] eq 27 (p. 20)
180 "movdqa %%xmm1, %%xmm3 \n\t" // r1:r0
181 "movdqa %%xmm2, %%xmm4 \n\t" // r3:r2
182 "psllq $1, %%xmm1 \n\t" // r1<<1:r0<<1
183 "psllq $1, %%xmm2 \n\t" // r3<<1:r2<<1
184 "psrlq $63, %%xmm3 \n\t" // r1>>63:r0>>63
185 "psrlq $63, %%xmm4 \n\t" // r3>>63:r2>>63
186 "movdqa %%xmm3, %%xmm5 \n\t" // r1>>63:r0>>63
187 "pslldq $8, %%xmm3 \n\t" // r0>>63:0
188 "pslldq $8, %%xmm4 \n\t" // r2>>63:0
189 "psrldq $8, %%xmm5 \n\t" // 0:r1>>63
190 "por %%xmm3, %%xmm1 \n\t" // r1<<1|r0>>63:r0<<1
191 "por %%xmm4, %%xmm2 \n\t" // r3<<1|r2>>62:r2<<1
192 "por %%xmm5, %%xmm2 \n\t" // r3<<1|r2>>62:r2<<1|r1>>63
195 * Now reduce modulo the GCM polynomial x^128 + x^7 + x^2 + x + 1
196 * using [CLMUL-WP] algorithm 5 (p. 20).
197 * Currently xmm2:xmm1 holds x3:x2:x1:x0 (already shifted).
200 "movdqa %%xmm1, %%xmm3 \n\t" // x1:x0
201 "movdqa %%xmm1, %%xmm4 \n\t" // same
202 "movdqa %%xmm1, %%xmm5 \n\t" // same
203 "psllq $63, %%xmm3 \n\t" // x1<<63:x0<<63 = stuff:a
204 "psllq $62, %%xmm4 \n\t" // x1<<62:x0<<62 = stuff:b
205 "psllq $57, %%xmm5 \n\t" // x1<<57:x0<<57 = stuff:c
208 "pxor %%xmm4, %%xmm3 \n\t" // stuff:a+b
209 "pxor %%xmm5, %%xmm3 \n\t" // stuff:a+b+c
210 "pslldq $8, %%xmm3 \n\t" // a+b+c:0
211 "pxor %%xmm3, %%xmm1 \n\t" // x1+a+b+c:x0 = d:x0
214 "movdqa %%xmm1,%%xmm0 \n\t" // d:x0
215 "movdqa %%xmm1,%%xmm4 \n\t" // same
216 "movdqa %%xmm1,%%xmm5 \n\t" // same
217 "psrlq $1, %%xmm0 \n\t" // e1:x0>>1 = e1:e0'
218 "psrlq $2, %%xmm4 \n\t" // f1:x0>>2 = f1:f0'
219 "psrlq $7, %%xmm5 \n\t" // g1:x0>>7 = g1:g0'
220 "pxor %%xmm4, %%xmm0 \n\t" // e1+f1:e0'+f0'
221 "pxor %%xmm5, %%xmm0 \n\t" // e1+f1+g1:e0'+f0'+g0'
222 // e0'+f0'+g0' is almost e0+f0+g0, ex\tcept for some missing
223 // bits carried from d. Now get those\t bits back in.
224 "movdqa %%xmm1,%%xmm3 \n\t" // d:x0
225 "movdqa %%xmm1,%%xmm4 \n\t" // same
226 "movdqa %%xmm1,%%xmm5 \n\t" // same
227 "psllq $63, %%xmm3 \n\t" // d<<63:stuff
228 "psllq $62, %%xmm4 \n\t" // d<<62:stuff
229 "psllq $57, %%xmm5 \n\t" // d<<57:stuff
230 "pxor %%xmm4, %%xmm3 \n\t" // d<<63+d<<62:stuff
231 "pxor %%xmm5, %%xmm3 \n\t" // missing bits of d:stuff
232 "psrldq $8, %%xmm3 \n\t" // 0:missing bits of d
233 "pxor %%xmm3, %%xmm0 \n\t" // e1+f1+g1:e0+f0+g0
234 "pxor %%xmm1, %%xmm0 \n\t" // h1:h0
235 "pxor %%xmm2, %%xmm0 \n\t" // x3+h1:x2+h0
237 "movdqu %%xmm0, (%2) \n\t" // done
239 : "r" (aa
), "r" (bb
), "r" (cc
)
240 : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" );
242 /* Now byte-reverse the outputs */
243 for( i
= 0; i
< 16; i
++ )
250 * Compute decryption round keys from encryption round keys
252 void mbedtls_aesni_inverse_key( unsigned char *invkey
,
253 const unsigned char *fwdkey
, int nr
)
255 unsigned char *ik
= invkey
;
256 const unsigned char *fk
= fwdkey
+ 16 * nr
;
258 memcpy( ik
, fk
, 16 );
260 for( fk
-= 16, ik
+= 16; fk
> fwdkey
; fk
-= 16, ik
+= 16 )
261 asm( "movdqu (%0), %%xmm0 \n\t"
262 AESIMC xmm0_xmm0
"\n\t"
263 "movdqu %%xmm0, (%1) \n\t"
266 : "memory", "xmm0" );
268 memcpy( ik
, fk
, 16 );
272 * Key expansion, 128-bit case
274 static void aesni_setkey_enc_128( unsigned char *rk
,
275 const unsigned char *key
)
277 asm( "movdqu (%1), %%xmm0 \n\t" // copy the original key
278 "movdqu %%xmm0, (%0) \n\t" // as round key 0
279 "jmp 2f \n\t" // skip auxiliary routine
282 * Finish generating the next round key.
284 * On entry xmm0 is r3:r2:r1:r0 and xmm1 is X:stuff:stuff:stuff
285 * with X = rot( sub( r3 ) ) ^ RCON.
287 * On exit, xmm0 is r7:r6:r5:r4
288 * with r4 = X + r0, r5 = r4 + r1, r6 = r5 + r2, r7 = r6 + r3
289 * and those are written to the round key buffer.
292 "pshufd $0xff, %%xmm1, %%xmm1 \n\t" // X:X:X:X
293 "pxor %%xmm0, %%xmm1 \n\t" // X+r3:X+r2:X+r1:r4
294 "pslldq $4, %%xmm0 \n\t" // r2:r1:r0:0
295 "pxor %%xmm0, %%xmm1 \n\t" // X+r3+r2:X+r2+r1:r5:r4
296 "pslldq $4, %%xmm0 \n\t" // etc
297 "pxor %%xmm0, %%xmm1 \n\t"
298 "pslldq $4, %%xmm0 \n\t"
299 "pxor %%xmm1, %%xmm0 \n\t" // update xmm0 for next time!
300 "add $16, %0 \n\t" // point to next round key
301 "movdqu %%xmm0, (%0) \n\t" // write it
306 AESKEYGENA xmm0_xmm1
",0x01 \n\tcall 1b \n\t"
307 AESKEYGENA xmm0_xmm1
",0x02 \n\tcall 1b \n\t"
308 AESKEYGENA xmm0_xmm1
",0x04 \n\tcall 1b \n\t"
309 AESKEYGENA xmm0_xmm1
",0x08 \n\tcall 1b \n\t"
310 AESKEYGENA xmm0_xmm1
",0x10 \n\tcall 1b \n\t"
311 AESKEYGENA xmm0_xmm1
",0x20 \n\tcall 1b \n\t"
312 AESKEYGENA xmm0_xmm1
",0x40 \n\tcall 1b \n\t"
313 AESKEYGENA xmm0_xmm1
",0x80 \n\tcall 1b \n\t"
314 AESKEYGENA xmm0_xmm1
",0x1B \n\tcall 1b \n\t"
315 AESKEYGENA xmm0_xmm1
",0x36 \n\tcall 1b \n\t"
317 : "r" (rk
), "r" (key
)
318 : "memory", "cc", "0" );
322 * Key expansion, 192-bit case
324 static void aesni_setkey_enc_192( unsigned char *rk
,
325 const unsigned char *key
)
327 asm( "movdqu (%1), %%xmm0 \n\t" // copy original round key
328 "movdqu %%xmm0, (%0) \n\t"
330 "movq 16(%1), %%xmm1 \n\t"
331 "movq %%xmm1, (%0) \n\t"
333 "jmp 2f \n\t" // skip auxiliary routine
336 * Finish generating the next 6 quarter-keys.
338 * On entry xmm0 is r3:r2:r1:r0, xmm1 is stuff:stuff:r5:r4
339 * and xmm2 is stuff:stuff:X:stuff with X = rot( sub( r3 ) ) ^ RCON.
341 * On exit, xmm0 is r9:r8:r7:r6 and xmm1 is stuff:stuff:r11:r10
342 * and those are written to the round key buffer.
345 "pshufd $0x55, %%xmm2, %%xmm2 \n\t" // X:X:X:X
346 "pxor %%xmm0, %%xmm2 \n\t" // X+r3:X+r2:X+r1:r4
347 "pslldq $4, %%xmm0 \n\t" // etc
348 "pxor %%xmm0, %%xmm2 \n\t"
349 "pslldq $4, %%xmm0 \n\t"
350 "pxor %%xmm0, %%xmm2 \n\t"
351 "pslldq $4, %%xmm0 \n\t"
352 "pxor %%xmm2, %%xmm0 \n\t" // update xmm0 = r9:r8:r7:r6
353 "movdqu %%xmm0, (%0) \n\t"
355 "pshufd $0xff, %%xmm0, %%xmm2 \n\t" // r9:r9:r9:r9
356 "pxor %%xmm1, %%xmm2 \n\t" // stuff:stuff:r9+r5:r10
357 "pslldq $4, %%xmm1 \n\t" // r2:r1:r0:0
358 "pxor %%xmm2, %%xmm1 \n\t" // xmm1 = stuff:stuff:r11:r10
359 "movq %%xmm1, (%0) \n\t"
364 AESKEYGENA xmm1_xmm2
",0x01 \n\tcall 1b \n\t"
365 AESKEYGENA xmm1_xmm2
",0x02 \n\tcall 1b \n\t"
366 AESKEYGENA xmm1_xmm2
",0x04 \n\tcall 1b \n\t"
367 AESKEYGENA xmm1_xmm2
",0x08 \n\tcall 1b \n\t"
368 AESKEYGENA xmm1_xmm2
",0x10 \n\tcall 1b \n\t"
369 AESKEYGENA xmm1_xmm2
",0x20 \n\tcall 1b \n\t"
370 AESKEYGENA xmm1_xmm2
",0x40 \n\tcall 1b \n\t"
371 AESKEYGENA xmm1_xmm2
",0x80 \n\tcall 1b \n\t"
374 : "r" (rk
), "r" (key
)
375 : "memory", "cc", "0" );
379 * Key expansion, 256-bit case
381 static void aesni_setkey_enc_256( unsigned char *rk
,
382 const unsigned char *key
)
384 asm( "movdqu (%1), %%xmm0 \n\t"
385 "movdqu %%xmm0, (%0) \n\t"
387 "movdqu 16(%1), %%xmm1 \n\t"
388 "movdqu %%xmm1, (%0) \n\t"
389 "jmp 2f \n\t" // skip auxiliary routine
392 * Finish generating the next two round keys.
394 * On entry xmm0 is r3:r2:r1:r0, xmm1 is r7:r6:r5:r4 and
395 * xmm2 is X:stuff:stuff:stuff with X = rot( sub( r7 )) ^ RCON
397 * On exit, xmm0 is r11:r10:r9:r8 and xmm1 is r15:r14:r13:r12
398 * and those have been written to the output buffer.
401 "pshufd $0xff, %%xmm2, %%xmm2 \n\t"
402 "pxor %%xmm0, %%xmm2 \n\t"
403 "pslldq $4, %%xmm0 \n\t"
404 "pxor %%xmm0, %%xmm2 \n\t"
405 "pslldq $4, %%xmm0 \n\t"
406 "pxor %%xmm0, %%xmm2 \n\t"
407 "pslldq $4, %%xmm0 \n\t"
408 "pxor %%xmm2, %%xmm0 \n\t"
410 "movdqu %%xmm0, (%0) \n\t"
412 /* Set xmm2 to stuff:Y:stuff:stuff with Y = subword( r11 )
413 * and proceed to generate next round key from there */
414 AESKEYGENA xmm0_xmm2
",0x00 \n\t"
415 "pshufd $0xaa, %%xmm2, %%xmm2 \n\t"
416 "pxor %%xmm1, %%xmm2 \n\t"
417 "pslldq $4, %%xmm1 \n\t"
418 "pxor %%xmm1, %%xmm2 \n\t"
419 "pslldq $4, %%xmm1 \n\t"
420 "pxor %%xmm1, %%xmm2 \n\t"
421 "pslldq $4, %%xmm1 \n\t"
422 "pxor %%xmm2, %%xmm1 \n\t"
424 "movdqu %%xmm1, (%0) \n\t"
428 * Main "loop" - Generating one more key than necessary,
429 * see definition of mbedtls_aes_context.buf
432 AESKEYGENA xmm1_xmm2
",0x01 \n\tcall 1b \n\t"
433 AESKEYGENA xmm1_xmm2
",0x02 \n\tcall 1b \n\t"
434 AESKEYGENA xmm1_xmm2
",0x04 \n\tcall 1b \n\t"
435 AESKEYGENA xmm1_xmm2
",0x08 \n\tcall 1b \n\t"
436 AESKEYGENA xmm1_xmm2
",0x10 \n\tcall 1b \n\t"
437 AESKEYGENA xmm1_xmm2
",0x20 \n\tcall 1b \n\t"
438 AESKEYGENA xmm1_xmm2
",0x40 \n\tcall 1b \n\t"
440 : "r" (rk
), "r" (key
)
441 : "memory", "cc", "0" );
445 * Key expansion, wrapper
447 int mbedtls_aesni_setkey_enc( unsigned char *rk
,
448 const unsigned char *key
,
453 case 128: aesni_setkey_enc_128( rk
, key
); break;
454 case 192: aesni_setkey_enc_192( rk
, key
); break;
455 case 256: aesni_setkey_enc_256( rk
, key
); break;
456 default : return( MBEDTLS_ERR_AES_INVALID_KEY_LENGTH
);
462 #endif /* MBEDTLS_HAVE_X86_64 */
464 #endif /* MBEDTLS_AESNI_C */