1 /*
2 * AES-NI support functions
3 *
4 * Copyright The Mbed TLS Contributors
5 * SPDX-License-Identifier: Apache-2.0 OR GPL-2.0-or-later
6 */
7
8 /*
9 * [AES-WP] https://www.intel.com/content/www/us/en/developer/articles/tool/intel-advanced-encryption-standard-aes-instructions-set.html
10 * [CLMUL-WP] https://www.intel.com/content/www/us/en/develop/download/intel-carry-less-multiplication-instruction-and-its-usage-for-computing-the-gcm-mode.html
11 */
12
13 #include "common.h"
14
15 #if defined(MBEDTLS_AESNI_C)
16
17 #include "aesni.h"
18
19 #include <string.h>
20
21 #if defined(MBEDTLS_AESNI_HAVE_CODE)
22
23 #if MBEDTLS_AESNI_HAVE_CODE == 2
24 #if defined(__GNUC__)
25 #include <cpuid.h>
26 #elif defined(_MSC_VER)
27 #include <intrin.h>
28 #else
29 #error "`__cpuid` required by MBEDTLS_AESNI_C is not supported by the compiler"
30 #endif
31 #include <immintrin.h>
32 #endif
33
34 #if defined(MBEDTLS_ARCH_IS_X86)
35 #if defined(MBEDTLS_COMPILER_IS_GCC)
36 #pragma GCC push_options
37 #pragma GCC target ("pclmul,sse2,aes")
38 #define MBEDTLS_POP_TARGET_PRAGMA
39 #elif defined(__clang__) && (__clang_major__ >= 5)
40 #pragma clang attribute push (__attribute__((target("pclmul,sse2,aes"))), apply_to=function)
41 #define MBEDTLS_POP_TARGET_PRAGMA
42 #endif
43 #endif
44
45 #if !defined(MBEDTLS_AES_USE_HARDWARE_ONLY)
46 /*
47 * AES-NI support detection routine
48 */
mbedtls_aesni_has_support(unsigned int what)49 int mbedtls_aesni_has_support(unsigned int what)
50 {
51 static int done = 0;
52 static unsigned int c = 0;
53
54 if (!done) {
55 #if MBEDTLS_AESNI_HAVE_CODE == 2
56 static int info[4] = { 0, 0, 0, 0 };
57 #if defined(_MSC_VER)
58 __cpuid(info, 1);
59 #else
60 __cpuid(1, info[0], info[1], info[2], info[3]);
61 #endif
62 c = info[2];
63 #else /* AESNI using asm */
64 asm ("movl $1, %%eax \n\t"
65 "cpuid \n\t"
66 : "=c" (c)
67 :
68 : "eax", "ebx", "edx");
69 #endif /* MBEDTLS_AESNI_HAVE_CODE */
70 done = 1;
71 }
72
73 return (c & what) != 0;
74 }
75 #endif /* !MBEDTLS_AES_USE_HARDWARE_ONLY */
76
77 #if MBEDTLS_AESNI_HAVE_CODE == 2
78
79 /*
80 * AES-NI AES-ECB block en(de)cryption
81 */
mbedtls_aesni_crypt_ecb(mbedtls_aes_context * ctx,int mode,const unsigned char input[16],unsigned char output[16])82 int mbedtls_aesni_crypt_ecb(mbedtls_aes_context *ctx,
83 int mode,
84 const unsigned char input[16],
85 unsigned char output[16])
86 {
87 const __m128i *rk = (const __m128i *) (ctx->buf + ctx->rk_offset);
88 unsigned nr = ctx->nr; // Number of remaining rounds
89
90 // Load round key 0
91 __m128i state;
92 memcpy(&state, input, 16);
93 state = _mm_xor_si128(state, rk[0]); // state ^= *rk;
94 ++rk;
95 --nr;
96
97 #if !defined(MBEDTLS_BLOCK_CIPHER_NO_DECRYPT)
98 if (mode == MBEDTLS_AES_DECRYPT) {
99 while (nr != 0) {
100 state = _mm_aesdec_si128(state, *rk);
101 ++rk;
102 --nr;
103 }
104 state = _mm_aesdeclast_si128(state, *rk);
105 } else
106 #else
107 (void) mode;
108 #endif
109 {
110 while (nr != 0) {
111 state = _mm_aesenc_si128(state, *rk);
112 ++rk;
113 --nr;
114 }
115 state = _mm_aesenclast_si128(state, *rk);
116 }
117
118 memcpy(output, &state, 16);
119 return 0;
120 }
121
122 /*
123 * GCM multiplication: c = a times b in GF(2^128)
124 * Based on [CLMUL-WP] algorithms 1 (with equation 27) and 5.
125 */
126
gcm_clmul(const __m128i aa,const __m128i bb,__m128i * cc,__m128i * dd)127 static void gcm_clmul(const __m128i aa, const __m128i bb,
128 __m128i *cc, __m128i *dd)
129 {
130 /*
131 * Caryless multiplication dd:cc = aa * bb
132 * using [CLMUL-WP] algorithm 1 (p. 12).
133 */
134 *cc = _mm_clmulepi64_si128(aa, bb, 0x00); // a0*b0 = c1:c0
135 *dd = _mm_clmulepi64_si128(aa, bb, 0x11); // a1*b1 = d1:d0
136 __m128i ee = _mm_clmulepi64_si128(aa, bb, 0x10); // a0*b1 = e1:e0
137 __m128i ff = _mm_clmulepi64_si128(aa, bb, 0x01); // a1*b0 = f1:f0
138 ff = _mm_xor_si128(ff, ee); // e1+f1:e0+f0
139 ee = ff; // e1+f1:e0+f0
140 ff = _mm_srli_si128(ff, 8); // 0:e1+f1
141 ee = _mm_slli_si128(ee, 8); // e0+f0:0
142 *dd = _mm_xor_si128(*dd, ff); // d1:d0+e1+f1
143 *cc = _mm_xor_si128(*cc, ee); // c1+e0+f0:c0
144 }
145
gcm_shift(__m128i * cc,__m128i * dd)146 static void gcm_shift(__m128i *cc, __m128i *dd)
147 {
148 /* [CMUCL-WP] Algorithm 5 Step 1: shift cc:dd one bit to the left,
149 * taking advantage of [CLMUL-WP] eq 27 (p. 18). */
150 // // *cc = r1:r0
151 // // *dd = r3:r2
152 __m128i cc_lo = _mm_slli_epi64(*cc, 1); // r1<<1:r0<<1
153 __m128i dd_lo = _mm_slli_epi64(*dd, 1); // r3<<1:r2<<1
154 __m128i cc_hi = _mm_srli_epi64(*cc, 63); // r1>>63:r0>>63
155 __m128i dd_hi = _mm_srli_epi64(*dd, 63); // r3>>63:r2>>63
156 __m128i xmm5 = _mm_srli_si128(cc_hi, 8); // 0:r1>>63
157 cc_hi = _mm_slli_si128(cc_hi, 8); // r0>>63:0
158 dd_hi = _mm_slli_si128(dd_hi, 8); // 0:r1>>63
159
160 *cc = _mm_or_si128(cc_lo, cc_hi); // r1<<1|r0>>63:r0<<1
161 *dd = _mm_or_si128(_mm_or_si128(dd_lo, dd_hi), xmm5); // r3<<1|r2>>62:r2<<1|r1>>63
162 }
163
gcm_reduce(__m128i xx)164 static __m128i gcm_reduce(__m128i xx)
165 {
166 // // xx = x1:x0
167 /* [CLMUL-WP] Algorithm 5 Step 2 */
168 __m128i aa = _mm_slli_epi64(xx, 63); // x1<<63:x0<<63 = stuff:a
169 __m128i bb = _mm_slli_epi64(xx, 62); // x1<<62:x0<<62 = stuff:b
170 __m128i cc = _mm_slli_epi64(xx, 57); // x1<<57:x0<<57 = stuff:c
171 __m128i dd = _mm_slli_si128(_mm_xor_si128(_mm_xor_si128(aa, bb), cc), 8); // a+b+c:0
172 return _mm_xor_si128(dd, xx); // x1+a+b+c:x0 = d:x0
173 }
174
gcm_mix(__m128i dx)175 static __m128i gcm_mix(__m128i dx)
176 {
177 /* [CLMUL-WP] Algorithm 5 Steps 3 and 4 */
178 __m128i ee = _mm_srli_epi64(dx, 1); // e1:x0>>1 = e1:e0'
179 __m128i ff = _mm_srli_epi64(dx, 2); // f1:x0>>2 = f1:f0'
180 __m128i gg = _mm_srli_epi64(dx, 7); // g1:x0>>7 = g1:g0'
181
182 // e0'+f0'+g0' is almost e0+f0+g0, except for some missing
183 // bits carried from d. Now get those bits back in.
184 __m128i eh = _mm_slli_epi64(dx, 63); // d<<63:stuff
185 __m128i fh = _mm_slli_epi64(dx, 62); // d<<62:stuff
186 __m128i gh = _mm_slli_epi64(dx, 57); // d<<57:stuff
187 __m128i hh = _mm_srli_si128(_mm_xor_si128(_mm_xor_si128(eh, fh), gh), 8); // 0:missing bits of d
188
189 return _mm_xor_si128(_mm_xor_si128(_mm_xor_si128(_mm_xor_si128(ee, ff), gg), hh), dx);
190 }
191
mbedtls_aesni_gcm_mult(unsigned char c[16],const unsigned char a[16],const unsigned char b[16])192 void mbedtls_aesni_gcm_mult(unsigned char c[16],
193 const unsigned char a[16],
194 const unsigned char b[16])
195 {
196 __m128i aa = { 0 }, bb = { 0 }, cc, dd;
197
198 /* The inputs are in big-endian order, so byte-reverse them */
199 for (size_t i = 0; i < 16; i++) {
200 ((uint8_t *) &aa)[i] = a[15 - i];
201 ((uint8_t *) &bb)[i] = b[15 - i];
202 }
203
204 gcm_clmul(aa, bb, &cc, &dd);
205 gcm_shift(&cc, &dd);
206 /*
207 * Now reduce modulo the GCM polynomial x^128 + x^7 + x^2 + x + 1
208 * using [CLMUL-WP] algorithm 5 (p. 18).
209 * Currently dd:cc holds x3:x2:x1:x0 (already shifted).
210 */
211 __m128i dx = gcm_reduce(cc);
212 __m128i xh = gcm_mix(dx);
213 cc = _mm_xor_si128(xh, dd); // x3+h1:x2+h0
214
215 /* Now byte-reverse the outputs */
216 for (size_t i = 0; i < 16; i++) {
217 c[i] = ((uint8_t *) &cc)[15 - i];
218 }
219
220 return;
221 }
222
223 /*
224 * Compute decryption round keys from encryption round keys
225 */
226 #if !defined(MBEDTLS_BLOCK_CIPHER_NO_DECRYPT)
mbedtls_aesni_inverse_key(unsigned char * invkey,const unsigned char * fwdkey,int nr)227 void mbedtls_aesni_inverse_key(unsigned char *invkey,
228 const unsigned char *fwdkey, int nr)
229 {
230 __m128i *ik = (__m128i *) invkey;
231 const __m128i *fk = (const __m128i *) fwdkey + nr;
232
233 *ik = *fk;
234 for (--fk, ++ik; fk > (const __m128i *) fwdkey; --fk, ++ik) {
235 *ik = _mm_aesimc_si128(*fk);
236 }
237 *ik = *fk;
238 }
239 #endif
240
241 /*
242 * Key expansion, 128-bit case
243 */
aesni_set_rk_128(__m128i state,__m128i xword)244 static __m128i aesni_set_rk_128(__m128i state, __m128i xword)
245 {
246 /*
247 * Finish generating the next round key.
248 *
249 * On entry state is r3:r2:r1:r0 and xword is X:stuff:stuff:stuff
250 * with X = rot( sub( r3 ) ) ^ RCON (obtained with AESKEYGENASSIST).
251 *
252 * On exit, xword is r7:r6:r5:r4
253 * with r4 = X + r0, r5 = r4 + r1, r6 = r5 + r2, r7 = r6 + r3
254 * and this is returned, to be written to the round key buffer.
255 */
256 xword = _mm_shuffle_epi32(xword, 0xff); // X:X:X:X
257 xword = _mm_xor_si128(xword, state); // X+r3:X+r2:X+r1:r4
258 state = _mm_slli_si128(state, 4); // r2:r1:r0:0
259 xword = _mm_xor_si128(xword, state); // X+r3+r2:X+r2+r1:r5:r4
260 state = _mm_slli_si128(state, 4); // r1:r0:0:0
261 xword = _mm_xor_si128(xword, state); // X+r3+r2+r1:r6:r5:r4
262 state = _mm_slli_si128(state, 4); // r0:0:0:0
263 state = _mm_xor_si128(xword, state); // r7:r6:r5:r4
264 return state;
265 }
266
aesni_setkey_enc_128(unsigned char * rk_bytes,const unsigned char * key)267 static void aesni_setkey_enc_128(unsigned char *rk_bytes,
268 const unsigned char *key)
269 {
270 __m128i *rk = (__m128i *) rk_bytes;
271
272 memcpy(&rk[0], key, 16);
273 rk[1] = aesni_set_rk_128(rk[0], _mm_aeskeygenassist_si128(rk[0], 0x01));
274 rk[2] = aesni_set_rk_128(rk[1], _mm_aeskeygenassist_si128(rk[1], 0x02));
275 rk[3] = aesni_set_rk_128(rk[2], _mm_aeskeygenassist_si128(rk[2], 0x04));
276 rk[4] = aesni_set_rk_128(rk[3], _mm_aeskeygenassist_si128(rk[3], 0x08));
277 rk[5] = aesni_set_rk_128(rk[4], _mm_aeskeygenassist_si128(rk[4], 0x10));
278 rk[6] = aesni_set_rk_128(rk[5], _mm_aeskeygenassist_si128(rk[5], 0x20));
279 rk[7] = aesni_set_rk_128(rk[6], _mm_aeskeygenassist_si128(rk[6], 0x40));
280 rk[8] = aesni_set_rk_128(rk[7], _mm_aeskeygenassist_si128(rk[7], 0x80));
281 rk[9] = aesni_set_rk_128(rk[8], _mm_aeskeygenassist_si128(rk[8], 0x1B));
282 rk[10] = aesni_set_rk_128(rk[9], _mm_aeskeygenassist_si128(rk[9], 0x36));
283 }
284
285 /*
286 * Key expansion, 192-bit case
287 */
288 #if !defined(MBEDTLS_AES_ONLY_128_BIT_KEY_LENGTH)
aesni_set_rk_192(__m128i * state0,__m128i * state1,__m128i xword,unsigned char * rk)289 static void aesni_set_rk_192(__m128i *state0, __m128i *state1, __m128i xword,
290 unsigned char *rk)
291 {
292 /*
293 * Finish generating the next 6 quarter-keys.
294 *
295 * On entry state0 is r3:r2:r1:r0, state1 is stuff:stuff:r5:r4
296 * and xword is stuff:stuff:X:stuff with X = rot( sub( r3 ) ) ^ RCON
297 * (obtained with AESKEYGENASSIST).
298 *
299 * On exit, state0 is r9:r8:r7:r6 and state1 is stuff:stuff:r11:r10
300 * and those are written to the round key buffer.
301 */
302 xword = _mm_shuffle_epi32(xword, 0x55); // X:X:X:X
303 xword = _mm_xor_si128(xword, *state0); // X+r3:X+r2:X+r1:X+r0
304 *state0 = _mm_slli_si128(*state0, 4); // r2:r1:r0:0
305 xword = _mm_xor_si128(xword, *state0); // X+r3+r2:X+r2+r1:X+r1+r0:X+r0
306 *state0 = _mm_slli_si128(*state0, 4); // r1:r0:0:0
307 xword = _mm_xor_si128(xword, *state0); // X+r3+r2+r1:X+r2+r1+r0:X+r1+r0:X+r0
308 *state0 = _mm_slli_si128(*state0, 4); // r0:0:0:0
309 xword = _mm_xor_si128(xword, *state0); // X+r3+r2+r1+r0:X+r2+r1+r0:X+r1+r0:X+r0
310 *state0 = xword; // = r9:r8:r7:r6
311
312 xword = _mm_shuffle_epi32(xword, 0xff); // r9:r9:r9:r9
313 xword = _mm_xor_si128(xword, *state1); // stuff:stuff:r9+r5:r9+r4
314 *state1 = _mm_slli_si128(*state1, 4); // stuff:stuff:r4:0
315 xword = _mm_xor_si128(xword, *state1); // stuff:stuff:r9+r5+r4:r9+r4
316 *state1 = xword; // = stuff:stuff:r11:r10
317
318 /* Store state0 and the low half of state1 into rk, which is conceptually
319 * an array of 24-byte elements. Since 24 is not a multiple of 16,
320 * rk is not necessarily aligned so just `*rk = *state0` doesn't work. */
321 memcpy(rk, state0, 16);
322 memcpy(rk + 16, state1, 8);
323 }
324
aesni_setkey_enc_192(unsigned char * rk,const unsigned char * key)325 static void aesni_setkey_enc_192(unsigned char *rk,
326 const unsigned char *key)
327 {
328 /* First round: use original key */
329 memcpy(rk, key, 24);
330 /* aes.c guarantees that rk is aligned on a 16-byte boundary. */
331 __m128i state0 = ((__m128i *) rk)[0];
332 __m128i state1 = _mm_loadl_epi64(((__m128i *) rk) + 1);
333
334 aesni_set_rk_192(&state0, &state1, _mm_aeskeygenassist_si128(state1, 0x01), rk + 24 * 1);
335 aesni_set_rk_192(&state0, &state1, _mm_aeskeygenassist_si128(state1, 0x02), rk + 24 * 2);
336 aesni_set_rk_192(&state0, &state1, _mm_aeskeygenassist_si128(state1, 0x04), rk + 24 * 3);
337 aesni_set_rk_192(&state0, &state1, _mm_aeskeygenassist_si128(state1, 0x08), rk + 24 * 4);
338 aesni_set_rk_192(&state0, &state1, _mm_aeskeygenassist_si128(state1, 0x10), rk + 24 * 5);
339 aesni_set_rk_192(&state0, &state1, _mm_aeskeygenassist_si128(state1, 0x20), rk + 24 * 6);
340 aesni_set_rk_192(&state0, &state1, _mm_aeskeygenassist_si128(state1, 0x40), rk + 24 * 7);
341 aesni_set_rk_192(&state0, &state1, _mm_aeskeygenassist_si128(state1, 0x80), rk + 24 * 8);
342 }
343 #endif /* !MBEDTLS_AES_ONLY_128_BIT_KEY_LENGTH */
344
345 /*
346 * Key expansion, 256-bit case
347 */
348 #if !defined(MBEDTLS_AES_ONLY_128_BIT_KEY_LENGTH)
aesni_set_rk_256(__m128i state0,__m128i state1,__m128i xword,__m128i * rk0,__m128i * rk1)349 static void aesni_set_rk_256(__m128i state0, __m128i state1, __m128i xword,
350 __m128i *rk0, __m128i *rk1)
351 {
352 /*
353 * Finish generating the next two round keys.
354 *
355 * On entry state0 is r3:r2:r1:r0, state1 is r7:r6:r5:r4 and
356 * xword is X:stuff:stuff:stuff with X = rot( sub( r7 )) ^ RCON
357 * (obtained with AESKEYGENASSIST).
358 *
359 * On exit, *rk0 is r11:r10:r9:r8 and *rk1 is r15:r14:r13:r12
360 */
361 xword = _mm_shuffle_epi32(xword, 0xff);
362 xword = _mm_xor_si128(xword, state0);
363 state0 = _mm_slli_si128(state0, 4);
364 xword = _mm_xor_si128(xword, state0);
365 state0 = _mm_slli_si128(state0, 4);
366 xword = _mm_xor_si128(xword, state0);
367 state0 = _mm_slli_si128(state0, 4);
368 state0 = _mm_xor_si128(state0, xword);
369 *rk0 = state0;
370
371 /* Set xword to stuff:Y:stuff:stuff with Y = subword( r11 )
372 * and proceed to generate next round key from there */
373 xword = _mm_aeskeygenassist_si128(state0, 0x00);
374 xword = _mm_shuffle_epi32(xword, 0xaa);
375 xword = _mm_xor_si128(xword, state1);
376 state1 = _mm_slli_si128(state1, 4);
377 xword = _mm_xor_si128(xword, state1);
378 state1 = _mm_slli_si128(state1, 4);
379 xword = _mm_xor_si128(xword, state1);
380 state1 = _mm_slli_si128(state1, 4);
381 state1 = _mm_xor_si128(state1, xword);
382 *rk1 = state1;
383 }
384
aesni_setkey_enc_256(unsigned char * rk_bytes,const unsigned char * key)385 static void aesni_setkey_enc_256(unsigned char *rk_bytes,
386 const unsigned char *key)
387 {
388 __m128i *rk = (__m128i *) rk_bytes;
389
390 memcpy(&rk[0], key, 16);
391 memcpy(&rk[1], key + 16, 16);
392
393 /*
394 * Main "loop" - Generating one more key than necessary,
395 * see definition of mbedtls_aes_context.buf
396 */
397 aesni_set_rk_256(rk[0], rk[1], _mm_aeskeygenassist_si128(rk[1], 0x01), &rk[2], &rk[3]);
398 aesni_set_rk_256(rk[2], rk[3], _mm_aeskeygenassist_si128(rk[3], 0x02), &rk[4], &rk[5]);
399 aesni_set_rk_256(rk[4], rk[5], _mm_aeskeygenassist_si128(rk[5], 0x04), &rk[6], &rk[7]);
400 aesni_set_rk_256(rk[6], rk[7], _mm_aeskeygenassist_si128(rk[7], 0x08), &rk[8], &rk[9]);
401 aesni_set_rk_256(rk[8], rk[9], _mm_aeskeygenassist_si128(rk[9], 0x10), &rk[10], &rk[11]);
402 aesni_set_rk_256(rk[10], rk[11], _mm_aeskeygenassist_si128(rk[11], 0x20), &rk[12], &rk[13]);
403 aesni_set_rk_256(rk[12], rk[13], _mm_aeskeygenassist_si128(rk[13], 0x40), &rk[14], &rk[15]);
404 }
405 #endif /* !MBEDTLS_AES_ONLY_128_BIT_KEY_LENGTH */
406
407 #if defined(MBEDTLS_POP_TARGET_PRAGMA)
408 #if defined(__clang__)
409 #pragma clang attribute pop
410 #elif defined(__GNUC__)
411 #pragma GCC pop_options
412 #endif
413 #undef MBEDTLS_POP_TARGET_PRAGMA
414 #endif
415
416 #else /* MBEDTLS_AESNI_HAVE_CODE == 1 */
417
418 #if defined(__has_feature)
419 #if __has_feature(memory_sanitizer)
420 #warning \
421 "MBEDTLS_AESNI_C is known to cause spurious error reports with some memory sanitizers as they do not understand the assembly code."
422 #endif
423 #endif
424
425 /*
426 * Binutils needs to be at least 2.19 to support AES-NI instructions.
427 * Unfortunately, a lot of users have a lower version now (2014-04).
428 * Emit bytecode directly in order to support "old" version of gas.
429 *
430 * Opcodes from the Intel architecture reference manual, vol. 3.
431 * We always use registers, so we don't need prefixes for memory operands.
432 * Operand macros are in gas order (src, dst) as opposed to Intel order
433 * (dst, src) in order to blend better into the surrounding assembly code.
434 */
435 #define AESDEC(regs) ".byte 0x66,0x0F,0x38,0xDE," regs "\n\t"
436 #define AESDECLAST(regs) ".byte 0x66,0x0F,0x38,0xDF," regs "\n\t"
437 #define AESENC(regs) ".byte 0x66,0x0F,0x38,0xDC," regs "\n\t"
438 #define AESENCLAST(regs) ".byte 0x66,0x0F,0x38,0xDD," regs "\n\t"
439 #define AESIMC(regs) ".byte 0x66,0x0F,0x38,0xDB," regs "\n\t"
440 #define AESKEYGENA(regs, imm) ".byte 0x66,0x0F,0x3A,0xDF," regs "," imm "\n\t"
441 #define PCLMULQDQ(regs, imm) ".byte 0x66,0x0F,0x3A,0x44," regs "," imm "\n\t"
442
443 #define xmm0_xmm0 "0xC0"
444 #define xmm0_xmm1 "0xC8"
445 #define xmm0_xmm2 "0xD0"
446 #define xmm0_xmm3 "0xD8"
447 #define xmm0_xmm4 "0xE0"
448 #define xmm1_xmm0 "0xC1"
449 #define xmm1_xmm2 "0xD1"
450
451 /*
452 * AES-NI AES-ECB block en(de)cryption
453 */
mbedtls_aesni_crypt_ecb(mbedtls_aes_context * ctx,int mode,const unsigned char input[16],unsigned char output[16])454 int mbedtls_aesni_crypt_ecb(mbedtls_aes_context *ctx,
455 int mode,
456 const unsigned char input[16],
457 unsigned char output[16])
458 {
459 asm ("movdqu (%3), %%xmm0 \n\t" // load input
460 "movdqu (%1), %%xmm1 \n\t" // load round key 0
461 "pxor %%xmm1, %%xmm0 \n\t" // round 0
462 "add $16, %1 \n\t" // point to next round key
463 "subl $1, %0 \n\t" // normal rounds = nr - 1
464 "test %2, %2 \n\t" // mode?
465 "jz 2f \n\t" // 0 = decrypt
466
467 "1: \n\t" // encryption loop
468 "movdqu (%1), %%xmm1 \n\t" // load round key
469 AESENC(xmm1_xmm0) // do round
470 "add $16, %1 \n\t" // point to next round key
471 "subl $1, %0 \n\t" // loop
472 "jnz 1b \n\t"
473 "movdqu (%1), %%xmm1 \n\t" // load round key
474 AESENCLAST(xmm1_xmm0) // last round
475 #if !defined(MBEDTLS_BLOCK_CIPHER_NO_DECRYPT)
476 "jmp 3f \n\t"
477
478 "2: \n\t" // decryption loop
479 "movdqu (%1), %%xmm1 \n\t"
480 AESDEC(xmm1_xmm0) // do round
481 "add $16, %1 \n\t"
482 "subl $1, %0 \n\t"
483 "jnz 2b \n\t"
484 "movdqu (%1), %%xmm1 \n\t" // load round key
485 AESDECLAST(xmm1_xmm0) // last round
486 #endif
487
488 "3: \n\t"
489 "movdqu %%xmm0, (%4) \n\t" // export output
490 :
491 : "r" (ctx->nr), "r" (ctx->buf + ctx->rk_offset), "r" (mode), "r" (input), "r" (output)
492 : "memory", "cc", "xmm0", "xmm1");
493
494
495 return 0;
496 }
497
498 /*
499 * GCM multiplication: c = a times b in GF(2^128)
500 * Based on [CLMUL-WP] algorithms 1 (with equation 27) and 5.
501 */
mbedtls_aesni_gcm_mult(unsigned char c[16],const unsigned char a[16],const unsigned char b[16])502 void mbedtls_aesni_gcm_mult(unsigned char c[16],
503 const unsigned char a[16],
504 const unsigned char b[16])
505 {
506 unsigned char aa[16], bb[16], cc[16];
507 size_t i;
508
509 /* The inputs are in big-endian order, so byte-reverse them */
510 for (i = 0; i < 16; i++) {
511 aa[i] = a[15 - i];
512 bb[i] = b[15 - i];
513 }
514
515 asm ("movdqu (%0), %%xmm0 \n\t" // a1:a0
516 "movdqu (%1), %%xmm1 \n\t" // b1:b0
517
518 /*
519 * Caryless multiplication xmm2:xmm1 = xmm0 * xmm1
520 * using [CLMUL-WP] algorithm 1 (p. 12).
521 */
522 "movdqa %%xmm1, %%xmm2 \n\t" // copy of b1:b0
523 "movdqa %%xmm1, %%xmm3 \n\t" // same
524 "movdqa %%xmm1, %%xmm4 \n\t" // same
525 PCLMULQDQ(xmm0_xmm1, "0x00") // a0*b0 = c1:c0
526 PCLMULQDQ(xmm0_xmm2, "0x11") // a1*b1 = d1:d0
527 PCLMULQDQ(xmm0_xmm3, "0x10") // a0*b1 = e1:e0
528 PCLMULQDQ(xmm0_xmm4, "0x01") // a1*b0 = f1:f0
529 "pxor %%xmm3, %%xmm4 \n\t" // e1+f1:e0+f0
530 "movdqa %%xmm4, %%xmm3 \n\t" // same
531 "psrldq $8, %%xmm4 \n\t" // 0:e1+f1
532 "pslldq $8, %%xmm3 \n\t" // e0+f0:0
533 "pxor %%xmm4, %%xmm2 \n\t" // d1:d0+e1+f1
534 "pxor %%xmm3, %%xmm1 \n\t" // c1+e0+f1:c0
535
536 /*
537 * Now shift the result one bit to the left,
538 * taking advantage of [CLMUL-WP] eq 27 (p. 18)
539 */
540 "movdqa %%xmm1, %%xmm3 \n\t" // r1:r0
541 "movdqa %%xmm2, %%xmm4 \n\t" // r3:r2
542 "psllq $1, %%xmm1 \n\t" // r1<<1:r0<<1
543 "psllq $1, %%xmm2 \n\t" // r3<<1:r2<<1
544 "psrlq $63, %%xmm3 \n\t" // r1>>63:r0>>63
545 "psrlq $63, %%xmm4 \n\t" // r3>>63:r2>>63
546 "movdqa %%xmm3, %%xmm5 \n\t" // r1>>63:r0>>63
547 "pslldq $8, %%xmm3 \n\t" // r0>>63:0
548 "pslldq $8, %%xmm4 \n\t" // r2>>63:0
549 "psrldq $8, %%xmm5 \n\t" // 0:r1>>63
550 "por %%xmm3, %%xmm1 \n\t" // r1<<1|r0>>63:r0<<1
551 "por %%xmm4, %%xmm2 \n\t" // r3<<1|r2>>62:r2<<1
552 "por %%xmm5, %%xmm2 \n\t" // r3<<1|r2>>62:r2<<1|r1>>63
553
554 /*
555 * Now reduce modulo the GCM polynomial x^128 + x^7 + x^2 + x + 1
556 * using [CLMUL-WP] algorithm 5 (p. 18).
557 * Currently xmm2:xmm1 holds x3:x2:x1:x0 (already shifted).
558 */
559 /* Step 2 (1) */
560 "movdqa %%xmm1, %%xmm3 \n\t" // x1:x0
561 "movdqa %%xmm1, %%xmm4 \n\t" // same
562 "movdqa %%xmm1, %%xmm5 \n\t" // same
563 "psllq $63, %%xmm3 \n\t" // x1<<63:x0<<63 = stuff:a
564 "psllq $62, %%xmm4 \n\t" // x1<<62:x0<<62 = stuff:b
565 "psllq $57, %%xmm5 \n\t" // x1<<57:x0<<57 = stuff:c
566
567 /* Step 2 (2) */
568 "pxor %%xmm4, %%xmm3 \n\t" // stuff:a+b
569 "pxor %%xmm5, %%xmm3 \n\t" // stuff:a+b+c
570 "pslldq $8, %%xmm3 \n\t" // a+b+c:0
571 "pxor %%xmm3, %%xmm1 \n\t" // x1+a+b+c:x0 = d:x0
572
573 /* Steps 3 and 4 */
574 "movdqa %%xmm1,%%xmm0 \n\t" // d:x0
575 "movdqa %%xmm1,%%xmm4 \n\t" // same
576 "movdqa %%xmm1,%%xmm5 \n\t" // same
577 "psrlq $1, %%xmm0 \n\t" // e1:x0>>1 = e1:e0'
578 "psrlq $2, %%xmm4 \n\t" // f1:x0>>2 = f1:f0'
579 "psrlq $7, %%xmm5 \n\t" // g1:x0>>7 = g1:g0'
580 "pxor %%xmm4, %%xmm0 \n\t" // e1+f1:e0'+f0'
581 "pxor %%xmm5, %%xmm0 \n\t" // e1+f1+g1:e0'+f0'+g0'
582 // e0'+f0'+g0' is almost e0+f0+g0, ex\tcept for some missing
583 // bits carried from d. Now get those\t bits back in.
584 "movdqa %%xmm1,%%xmm3 \n\t" // d:x0
585 "movdqa %%xmm1,%%xmm4 \n\t" // same
586 "movdqa %%xmm1,%%xmm5 \n\t" // same
587 "psllq $63, %%xmm3 \n\t" // d<<63:stuff
588 "psllq $62, %%xmm4 \n\t" // d<<62:stuff
589 "psllq $57, %%xmm5 \n\t" // d<<57:stuff
590 "pxor %%xmm4, %%xmm3 \n\t" // d<<63+d<<62:stuff
591 "pxor %%xmm5, %%xmm3 \n\t" // missing bits of d:stuff
592 "psrldq $8, %%xmm3 \n\t" // 0:missing bits of d
593 "pxor %%xmm3, %%xmm0 \n\t" // e1+f1+g1:e0+f0+g0
594 "pxor %%xmm1, %%xmm0 \n\t" // h1:h0
595 "pxor %%xmm2, %%xmm0 \n\t" // x3+h1:x2+h0
596
597 "movdqu %%xmm0, (%2) \n\t" // done
598 :
599 : "r" (aa), "r" (bb), "r" (cc)
600 : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5");
601
602 /* Now byte-reverse the outputs */
603 for (i = 0; i < 16; i++) {
604 c[i] = cc[15 - i];
605 }
606
607 return;
608 }
609
610 /*
611 * Compute decryption round keys from encryption round keys
612 */
613 #if !defined(MBEDTLS_BLOCK_CIPHER_NO_DECRYPT)
mbedtls_aesni_inverse_key(unsigned char * invkey,const unsigned char * fwdkey,int nr)614 void mbedtls_aesni_inverse_key(unsigned char *invkey,
615 const unsigned char *fwdkey, int nr)
616 {
617 unsigned char *ik = invkey;
618 const unsigned char *fk = fwdkey + 16 * nr;
619
620 memcpy(ik, fk, 16);
621
622 for (fk -= 16, ik += 16; fk > fwdkey; fk -= 16, ik += 16) {
623 asm ("movdqu (%0), %%xmm0 \n\t"
624 AESIMC(xmm0_xmm0)
625 "movdqu %%xmm0, (%1) \n\t"
626 :
627 : "r" (fk), "r" (ik)
628 : "memory", "xmm0");
629 }
630
631 memcpy(ik, fk, 16);
632 }
633 #endif
634
635 /*
636 * Key expansion, 128-bit case
637 */
aesni_setkey_enc_128(unsigned char * rk,const unsigned char * key)638 static void aesni_setkey_enc_128(unsigned char *rk,
639 const unsigned char *key)
640 {
641 asm ("movdqu (%1), %%xmm0 \n\t" // copy the original key
642 "movdqu %%xmm0, (%0) \n\t" // as round key 0
643 "jmp 2f \n\t" // skip auxiliary routine
644
645 /*
646 * Finish generating the next round key.
647 *
648 * On entry xmm0 is r3:r2:r1:r0 and xmm1 is X:stuff:stuff:stuff
649 * with X = rot( sub( r3 ) ) ^ RCON.
650 *
651 * On exit, xmm0 is r7:r6:r5:r4
652 * with r4 = X + r0, r5 = r4 + r1, r6 = r5 + r2, r7 = r6 + r3
653 * and those are written to the round key buffer.
654 */
655 "1: \n\t"
656 "pshufd $0xff, %%xmm1, %%xmm1 \n\t" // X:X:X:X
657 "pxor %%xmm0, %%xmm1 \n\t" // X+r3:X+r2:X+r1:r4
658 "pslldq $4, %%xmm0 \n\t" // r2:r1:r0:0
659 "pxor %%xmm0, %%xmm1 \n\t" // X+r3+r2:X+r2+r1:r5:r4
660 "pslldq $4, %%xmm0 \n\t" // etc
661 "pxor %%xmm0, %%xmm1 \n\t"
662 "pslldq $4, %%xmm0 \n\t"
663 "pxor %%xmm1, %%xmm0 \n\t" // update xmm0 for next time!
664 "add $16, %0 \n\t" // point to next round key
665 "movdqu %%xmm0, (%0) \n\t" // write it
666 "ret \n\t"
667
668 /* Main "loop" */
669 "2: \n\t"
670 AESKEYGENA(xmm0_xmm1, "0x01") "call 1b \n\t"
671 AESKEYGENA(xmm0_xmm1, "0x02") "call 1b \n\t"
672 AESKEYGENA(xmm0_xmm1, "0x04") "call 1b \n\t"
673 AESKEYGENA(xmm0_xmm1, "0x08") "call 1b \n\t"
674 AESKEYGENA(xmm0_xmm1, "0x10") "call 1b \n\t"
675 AESKEYGENA(xmm0_xmm1, "0x20") "call 1b \n\t"
676 AESKEYGENA(xmm0_xmm1, "0x40") "call 1b \n\t"
677 AESKEYGENA(xmm0_xmm1, "0x80") "call 1b \n\t"
678 AESKEYGENA(xmm0_xmm1, "0x1B") "call 1b \n\t"
679 AESKEYGENA(xmm0_xmm1, "0x36") "call 1b \n\t"
680 :
681 : "r" (rk), "r" (key)
682 : "memory", "cc", "0");
683 }
684
685 /*
686 * Key expansion, 192-bit case
687 */
688 #if !defined(MBEDTLS_AES_ONLY_128_BIT_KEY_LENGTH)
aesni_setkey_enc_192(unsigned char * rk,const unsigned char * key)689 static void aesni_setkey_enc_192(unsigned char *rk,
690 const unsigned char *key)
691 {
692 asm ("movdqu (%1), %%xmm0 \n\t" // copy original round key
693 "movdqu %%xmm0, (%0) \n\t"
694 "add $16, %0 \n\t"
695 "movq 16(%1), %%xmm1 \n\t"
696 "movq %%xmm1, (%0) \n\t"
697 "add $8, %0 \n\t"
698 "jmp 2f \n\t" // skip auxiliary routine
699
700 /*
701 * Finish generating the next 6 quarter-keys.
702 *
703 * On entry xmm0 is r3:r2:r1:r0, xmm1 is stuff:stuff:r5:r4
704 * and xmm2 is stuff:stuff:X:stuff with X = rot( sub( r3 ) ) ^ RCON.
705 *
706 * On exit, xmm0 is r9:r8:r7:r6 and xmm1 is stuff:stuff:r11:r10
707 * and those are written to the round key buffer.
708 */
709 "1: \n\t"
710 "pshufd $0x55, %%xmm2, %%xmm2 \n\t" // X:X:X:X
711 "pxor %%xmm0, %%xmm2 \n\t" // X+r3:X+r2:X+r1:r4
712 "pslldq $4, %%xmm0 \n\t" // etc
713 "pxor %%xmm0, %%xmm2 \n\t"
714 "pslldq $4, %%xmm0 \n\t"
715 "pxor %%xmm0, %%xmm2 \n\t"
716 "pslldq $4, %%xmm0 \n\t"
717 "pxor %%xmm2, %%xmm0 \n\t" // update xmm0 = r9:r8:r7:r6
718 "movdqu %%xmm0, (%0) \n\t"
719 "add $16, %0 \n\t"
720 "pshufd $0xff, %%xmm0, %%xmm2 \n\t" // r9:r9:r9:r9
721 "pxor %%xmm1, %%xmm2 \n\t" // stuff:stuff:r9+r5:r10
722 "pslldq $4, %%xmm1 \n\t" // r2:r1:r0:0
723 "pxor %%xmm2, %%xmm1 \n\t" // xmm1 = stuff:stuff:r11:r10
724 "movq %%xmm1, (%0) \n\t"
725 "add $8, %0 \n\t"
726 "ret \n\t"
727
728 "2: \n\t"
729 AESKEYGENA(xmm1_xmm2, "0x01") "call 1b \n\t"
730 AESKEYGENA(xmm1_xmm2, "0x02") "call 1b \n\t"
731 AESKEYGENA(xmm1_xmm2, "0x04") "call 1b \n\t"
732 AESKEYGENA(xmm1_xmm2, "0x08") "call 1b \n\t"
733 AESKEYGENA(xmm1_xmm2, "0x10") "call 1b \n\t"
734 AESKEYGENA(xmm1_xmm2, "0x20") "call 1b \n\t"
735 AESKEYGENA(xmm1_xmm2, "0x40") "call 1b \n\t"
736 AESKEYGENA(xmm1_xmm2, "0x80") "call 1b \n\t"
737
738 :
739 : "r" (rk), "r" (key)
740 : "memory", "cc", "0");
741 }
742 #endif /* !MBEDTLS_AES_ONLY_128_BIT_KEY_LENGTH */
743
744 /*
745 * Key expansion, 256-bit case
746 */
747 #if !defined(MBEDTLS_AES_ONLY_128_BIT_KEY_LENGTH)
aesni_setkey_enc_256(unsigned char * rk,const unsigned char * key)748 static void aesni_setkey_enc_256(unsigned char *rk,
749 const unsigned char *key)
750 {
751 asm ("movdqu (%1), %%xmm0 \n\t"
752 "movdqu %%xmm0, (%0) \n\t"
753 "add $16, %0 \n\t"
754 "movdqu 16(%1), %%xmm1 \n\t"
755 "movdqu %%xmm1, (%0) \n\t"
756 "jmp 2f \n\t" // skip auxiliary routine
757
758 /*
759 * Finish generating the next two round keys.
760 *
761 * On entry xmm0 is r3:r2:r1:r0, xmm1 is r7:r6:r5:r4 and
762 * xmm2 is X:stuff:stuff:stuff with X = rot( sub( r7 )) ^ RCON
763 *
764 * On exit, xmm0 is r11:r10:r9:r8 and xmm1 is r15:r14:r13:r12
765 * and those have been written to the output buffer.
766 */
767 "1: \n\t"
768 "pshufd $0xff, %%xmm2, %%xmm2 \n\t"
769 "pxor %%xmm0, %%xmm2 \n\t"
770 "pslldq $4, %%xmm0 \n\t"
771 "pxor %%xmm0, %%xmm2 \n\t"
772 "pslldq $4, %%xmm0 \n\t"
773 "pxor %%xmm0, %%xmm2 \n\t"
774 "pslldq $4, %%xmm0 \n\t"
775 "pxor %%xmm2, %%xmm0 \n\t"
776 "add $16, %0 \n\t"
777 "movdqu %%xmm0, (%0) \n\t"
778
779 /* Set xmm2 to stuff:Y:stuff:stuff with Y = subword( r11 )
780 * and proceed to generate next round key from there */
781 AESKEYGENA(xmm0_xmm2, "0x00")
782 "pshufd $0xaa, %%xmm2, %%xmm2 \n\t"
783 "pxor %%xmm1, %%xmm2 \n\t"
784 "pslldq $4, %%xmm1 \n\t"
785 "pxor %%xmm1, %%xmm2 \n\t"
786 "pslldq $4, %%xmm1 \n\t"
787 "pxor %%xmm1, %%xmm2 \n\t"
788 "pslldq $4, %%xmm1 \n\t"
789 "pxor %%xmm2, %%xmm1 \n\t"
790 "add $16, %0 \n\t"
791 "movdqu %%xmm1, (%0) \n\t"
792 "ret \n\t"
793
794 /*
795 * Main "loop" - Generating one more key than necessary,
796 * see definition of mbedtls_aes_context.buf
797 */
798 "2: \n\t"
799 AESKEYGENA(xmm1_xmm2, "0x01") "call 1b \n\t"
800 AESKEYGENA(xmm1_xmm2, "0x02") "call 1b \n\t"
801 AESKEYGENA(xmm1_xmm2, "0x04") "call 1b \n\t"
802 AESKEYGENA(xmm1_xmm2, "0x08") "call 1b \n\t"
803 AESKEYGENA(xmm1_xmm2, "0x10") "call 1b \n\t"
804 AESKEYGENA(xmm1_xmm2, "0x20") "call 1b \n\t"
805 AESKEYGENA(xmm1_xmm2, "0x40") "call 1b \n\t"
806 :
807 : "r" (rk), "r" (key)
808 : "memory", "cc", "0");
809 }
810 #endif /* !MBEDTLS_AES_ONLY_128_BIT_KEY_LENGTH */
811
812 #endif /* MBEDTLS_AESNI_HAVE_CODE */
813
814 /*
815 * Key expansion, wrapper
816 */
mbedtls_aesni_setkey_enc(unsigned char * rk,const unsigned char * key,size_t bits)817 int mbedtls_aesni_setkey_enc(unsigned char *rk,
818 const unsigned char *key,
819 size_t bits)
820 {
821 switch (bits) {
822 case 128: aesni_setkey_enc_128(rk, key); break;
823 #if !defined(MBEDTLS_AES_ONLY_128_BIT_KEY_LENGTH)
824 case 192: aesni_setkey_enc_192(rk, key); break;
825 case 256: aesni_setkey_enc_256(rk, key); break;
826 #endif /* !MBEDTLS_AES_ONLY_128_BIT_KEY_LENGTH */
827 default: return MBEDTLS_ERR_AES_INVALID_KEY_LENGTH;
828 }
829
830 return 0;
831 }
832
833 #endif /* MBEDTLS_AESNI_HAVE_CODE */
834
835 #endif /* MBEDTLS_AESNI_C */
836
