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  */
mbedtls_aesni_has_support(unsigned int what)48 int mbedtls_aesni_has_support( unsigned int what )
49 {
50     static int done = 0;
51     static unsigned int c = 0;
52 
53     if( ! done )
54     {
55         asm( "movl  $1, %%eax   \n\t"
56              "cpuid             \n\t"
57              : "=c" (c)
58              :
59              : "eax", "ebx", "edx" );
60         done = 1;
61     }
62 
63     return( ( c & what ) != 0 );
64 }
65 
66 /*
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.
70  *
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.
75  */
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,"
83 
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"
91 
92 /*
93  * AES-NI AES-ECB block en(de)cryption
94  */
mbedtls_aesni_crypt_ecb(mbedtls_aes_context * ctx,int mode,const unsigned char input[16],unsigned char output[16])95 int mbedtls_aesni_crypt_ecb( mbedtls_aes_context *ctx,
96                      int mode,
97                      const unsigned char input[16],
98                      unsigned char output[16] )
99 {
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          "add       $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
107 
108          "1:                        \n\t" // encryption loop
109          "movdqu    (%1), %%xmm1    \n\t" // load round key
110          AESENC     xmm1_xmm0      "\n\t" // do round
111          "add       $16, %1         \n\t" // point to next round key
112          "subl      $1, %0          \n\t" // loop
113          "jnz       1b              \n\t"
114          "movdqu    (%1), %%xmm1    \n\t" // load round key
115          AESENCLAST xmm1_xmm0      "\n\t" // last round
116          "jmp       3f              \n\t"
117 
118          "2:                        \n\t" // decryption loop
119          "movdqu    (%1), %%xmm1    \n\t"
120          AESDEC     xmm1_xmm0      "\n\t" // do round
121          "add       $16, %1         \n\t"
122          "subl      $1, %0          \n\t"
123          "jnz       2b              \n\t"
124          "movdqu    (%1), %%xmm1    \n\t" // load round key
125          AESDECLAST xmm1_xmm0      "\n\t" // last round
126 
127          "3:                        \n\t"
128          "movdqu    %%xmm0, (%4)    \n\t" // export output
129          :
130          : "r" (ctx->nr), "r" (ctx->rk), "r" (mode), "r" (input), "r" (output)
131          : "memory", "cc", "xmm0", "xmm1" );
132 
133 
134     return( 0 );
135 }
136 
137 /*
138  * GCM multiplication: c = a times b in GF(2^128)
139  * Based on [CLMUL-WP] algorithms 1 (with equation 27) and 5.
140  */
mbedtls_aesni_gcm_mult(unsigned char c[16],const unsigned char a[16],const unsigned char b[16])141 void mbedtls_aesni_gcm_mult( unsigned char c[16],
142                      const unsigned char a[16],
143                      const unsigned char b[16] )
144 {
145     unsigned char aa[16], bb[16], cc[16];
146     size_t i;
147 
148     /* The inputs are in big-endian order, so byte-reverse them */
149     for( i = 0; i < 16; i++ )
150     {
151         aa[i] = a[15 - i];
152         bb[i] = b[15 - i];
153     }
154 
155     asm( "movdqu (%0), %%xmm0               \n\t" // a1:a0
156          "movdqu (%1), %%xmm1               \n\t" // b1:b0
157 
158          /*
159           * Caryless multiplication xmm2:xmm1 = xmm0 * xmm1
160           * using [CLMUL-WP] algorithm 1 (p. 13).
161           */
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
175 
176          /*
177           * Now shift the result one bit to the left,
178           * taking advantage of [CLMUL-WP] eq 27 (p. 20)
179           */
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
193 
194          /*
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).
198           */
199          /* Step 2 (1) */
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
206 
207          /* Step 2 (2) */
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
212 
213          /* Steps 3 and 4 */
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
236 
237          "movdqu %%xmm0, (%2)               \n\t" // done
238          :
239          : "r" (aa), "r" (bb), "r" (cc)
240          : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" );
241 
242     /* Now byte-reverse the outputs */
243     for( i = 0; i < 16; i++ )
244         c[i] = cc[15 - i];
245 
246     return;
247 }
248 
249 /*
250  * Compute decryption round keys from encryption round keys
251  */
mbedtls_aesni_inverse_key(unsigned char * invkey,const unsigned char * fwdkey,int nr)252 void mbedtls_aesni_inverse_key( unsigned char *invkey,
253                         const unsigned char *fwdkey, int nr )
254 {
255     unsigned char *ik = invkey;
256     const unsigned char *fk = fwdkey + 16 * nr;
257 
258     memcpy( ik, fk, 16 );
259 
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"
264              :
265              : "r" (fk), "r" (ik)
266              : "memory", "xmm0" );
267 
268     memcpy( ik, fk, 16 );
269 }
270 
271 /*
272  * Key expansion, 128-bit case
273  */
aesni_setkey_enc_128(unsigned char * rk,const unsigned char * key)274 static void aesni_setkey_enc_128( unsigned char *rk,
275                                   const unsigned char *key )
276 {
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
280 
281          /*
282           * Finish generating the next round key.
283           *
284           * On entry xmm0 is r3:r2:r1:r0 and xmm1 is X:stuff:stuff:stuff
285           * with X = rot( sub( r3 ) ) ^ RCON.
286           *
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.
290           */
291          "1:                                \n\t"
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
302          "ret                               \n\t"
303 
304          /* Main "loop" */
305          "2:                                \n\t"
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"
316          :
317          : "r" (rk), "r" (key)
318          : "memory", "cc", "0" );
319 }
320 
321 /*
322  * Key expansion, 192-bit case
323  */
aesni_setkey_enc_192(unsigned char * rk,const unsigned char * key)324 static void aesni_setkey_enc_192( unsigned char *rk,
325                                   const unsigned char *key )
326 {
327     asm( "movdqu (%1), %%xmm0   \n\t" // copy original round key
328          "movdqu %%xmm0, (%0)   \n\t"
329          "add $16, %0           \n\t"
330          "movq 16(%1), %%xmm1   \n\t"
331          "movq %%xmm1, (%0)     \n\t"
332          "add $8, %0            \n\t"
333          "jmp 2f                \n\t" // skip auxiliary routine
334 
335          /*
336           * Finish generating the next 6 quarter-keys.
337           *
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.
340           *
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.
343           */
344          "1:                            \n\t"
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"
354          "add $16, %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"
360          "add $8, %0                    \n\t"
361          "ret                           \n\t"
362 
363          "2:                            \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"
372 
373          :
374          : "r" (rk), "r" (key)
375          : "memory", "cc", "0" );
376 }
377 
378 /*
379  * Key expansion, 256-bit case
380  */
aesni_setkey_enc_256(unsigned char * rk,const unsigned char * key)381 static void aesni_setkey_enc_256( unsigned char *rk,
382                                   const unsigned char *key )
383 {
384     asm( "movdqu (%1), %%xmm0           \n\t"
385          "movdqu %%xmm0, (%0)           \n\t"
386          "add $16, %0                   \n\t"
387          "movdqu 16(%1), %%xmm1         \n\t"
388          "movdqu %%xmm1, (%0)           \n\t"
389          "jmp 2f                        \n\t" // skip auxiliary routine
390 
391          /*
392           * Finish generating the next two round keys.
393           *
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
396           *
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.
399           */
400          "1:                                \n\t"
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"
409          "add $16, %0                       \n\t"
410          "movdqu %%xmm0, (%0)               \n\t"
411 
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"
423          "add $16, %0                       \n\t"
424          "movdqu %%xmm1, (%0)               \n\t"
425          "ret                               \n\t"
426 
427          /*
428           * Main "loop" - Generating one more key than necessary,
429           * see definition of mbedtls_aes_context.buf
430           */
431          "2:                                \n\t"
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"
439          :
440          : "r" (rk), "r" (key)
441          : "memory", "cc", "0" );
442 }
443 
444 /*
445  * Key expansion, wrapper
446  */
mbedtls_aesni_setkey_enc(unsigned char * rk,const unsigned char * key,size_t bits)447 int mbedtls_aesni_setkey_enc( unsigned char *rk,
448                       const unsigned char *key,
449                       size_t bits )
450 {
451     switch( bits )
452     {
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 );
457     }
458 
459     return( 0 );
460 }
461 
462 #endif /* MBEDTLS_HAVE_X86_64 */
463 
464 #endif /* MBEDTLS_AESNI_C */
465