Commit | Line | Data |
---|---|---|
1a0c09c4 MW |
1 | /// -*- mode: asm; asm-comment-char: ?/ -*- |
2 | /// | |
3 | /// AESNI-based implementation of Rijndael | |
4 | /// | |
5 | /// (c) 2015 Straylight/Edgeware | |
6 | /// | |
7 | ||
8 | ///----- Licensing notice --------------------------------------------------- | |
9 | /// | |
10 | /// This file is part of Catacomb. | |
11 | /// | |
12 | /// Catacomb is free software; you can redistribute it and/or modify | |
13 | /// it under the terms of the GNU Library General Public License as | |
14 | /// published by the Free Software Foundation; either version 2 of the | |
15 | /// License, or (at your option) any later version. | |
16 | /// | |
17 | /// Catacomb is distributed in the hope that it will be useful, | |
18 | /// but WITHOUT ANY WARRANTY; without even the implied warranty of | |
19 | /// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
20 | /// GNU Library General Public License for more details. | |
21 | /// | |
22 | /// You should have received a copy of the GNU Library General Public | |
23 | /// License along with Catacomb; if not, write to the Free | |
24 | /// Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, | |
25 | /// MA 02111-1307, USA. | |
26 | ||
27 | ///-------------------------------------------------------------------------- | |
28 | /// External definitions. | |
29 | ||
30 | #include "config.h" | |
31 | #include "asm-common.h" | |
32 | ||
33 | ///-------------------------------------------------------------------------- | |
34 | /// External definitions. | |
35 | ||
36 | .globl F(abort) | |
37 | .globl F(rijndael_rcon) | |
38 | ||
47103664 MW |
39 | ///-------------------------------------------------------------------------- |
40 | /// Local utilities. | |
41 | ||
42 | // Magic constants for shuffling. | |
43 | #define ROTL 0x93 | |
44 | #define ROT2 0x4e | |
45 | #define ROTR 0x39 | |
46 | ||
1a0c09c4 MW |
47 | ///-------------------------------------------------------------------------- |
48 | /// Main code. | |
49 | ||
50 | .arch .aes | |
bc9ac7eb | 51 | .text |
1a0c09c4 MW |
52 | |
53 | /// The AESNI instructions implement a little-endian version of AES, but | |
54 | /// Catacomb's internal interface presents as big-endian so as to work better | |
55 | /// with things like GCM. We therefore maintain the round keys in | |
56 | /// little-endian form, and have to end-swap blocks in and out. | |
57 | /// | |
58 | /// For added amusement, the AESNI instructions don't implement the | |
59 | /// larger-block versions of Rijndael, so we have to end-swap the keys if | |
60 | /// we're preparing for one of those. | |
61 | ||
62 | // Useful constants. | |
63 | .equ maxrounds, 16 // maximum number of rounds | |
64 | .equ maxblksz, 32 // maximum block size, in bytes | |
65 | .equ kbufsz, maxblksz*(maxrounds + 1) // size of a key-schedule buffer | |
66 | ||
67 | // Context structure. | |
68 | .equ nr, 0 // number of rounds | |
69 | .equ w, nr + 4 // encryption key words | |
70 | .equ wi, w + kbufsz // decryption key words | |
71 | ||
72 | ///-------------------------------------------------------------------------- | |
73 | /// Key setup. | |
74 | ||
0f23f75f | 75 | FUNC(rijndael_setup_x86ish_aesni) |
1a0c09c4 | 76 | |
0f23f75f MW |
77 | #if CPUFAM_X86 |
78 | // Arguments are on the stack. We'll need to stack the caller's | |
79 | // register veriables, but we'll manage. | |
1a0c09c4 | 80 | |
0f23f75f MW |
81 | # define CTX ebp // context pointer |
82 | # define BLKSZ [esp + 24] // block size | |
83 | ||
84 | # define SI esi // source pointer | |
85 | # define DI edi // destination pointer | |
86 | ||
87 | # define KSZ ebx // key size | |
88 | # define KSZo ebx // ... as address offset | |
89 | # define NKW edx // total number of key words | |
90 | # define NKW_NEEDS_REFRESH 1 // ... needs recalculating | |
91 | # define RCON ecx // round constants table | |
92 | # define LIM edx // limit pointer | |
93 | # define LIMn edx // ... as integer offset from base | |
94 | ||
95 | # define NR ecx // number of rounds | |
96 | # define LRK eax // distance to last key | |
97 | # define LRKo eax // ... as address offset | |
98 | # define BLKOFF edx // block size in bytes | |
99 | # define BLKOFFo edx // ... as address offset | |
100 | ||
101 | // Stack the caller's registers. | |
1a0c09c4 MW |
102 | push ebp |
103 | push ebx | |
104 | push esi | |
105 | push edi | |
106 | ||
0f23f75f MW |
107 | // Set up our own variables. |
108 | mov CTX, [esp + 20] // context base pointer | |
109 | mov SI, [esp + 28] // key material | |
110 | mov KSZ, [esp + 32] // key size, in words | |
111 | #endif | |
112 | ||
113 | #if CPUFAM_AMD64 && ABI_SYSV | |
114 | // Arguments are in registers. We have plenty, but, to be honest, | |
115 | // the initial register allocation is a bit annoying. | |
116 | ||
117 | # define CTX r8 // context pointer | |
118 | # define BLKSZ r9d // block size | |
119 | ||
120 | # define SI rsi // source pointer | |
121 | # define DI rdi // destination pointer | |
122 | ||
123 | # define KSZ edx // key size | |
124 | # define KSZo rdx // ... as address offset | |
125 | # define NKW r10d // total number of key words | |
126 | # define RCON rdi // round constants table | |
127 | # define LIMn ecx // limit pointer | |
128 | # define LIM rcx // ... as integer offset from base | |
129 | ||
130 | # define NR ecx // number of rounds | |
131 | # define LRK eax // distance to last key | |
132 | # define LRKo rax // ... as address offset | |
133 | # define BLKOFF r9d // block size in bytes | |
134 | # define BLKOFFo r9 // ... as address offset | |
135 | ||
136 | // Move arguments to more useful places. | |
137 | mov CTX, rdi // context base pointer | |
138 | mov BLKSZ, esi // block size in words | |
139 | mov SI, rdx // key material | |
140 | mov KSZ, ecx // key size, in words | |
141 | #endif | |
142 | ||
143 | #if CPUFAM_AMD64 && ABI_WIN | |
144 | // Arguments are in different registers, and they're a little tight. | |
145 | ||
146 | # define CTX r8 // context pointer | |
147 | # define BLKSZ edx // block size | |
148 | ||
149 | # define SI rsi // source pointer | |
150 | # define DI rdi // destination pointer | |
151 | ||
152 | # define KSZ r9d // key size | |
153 | # define KSZo r9 // ... as address offset | |
154 | # define NKW r10d // total number of key words | |
155 | # define RCON rdi // round constants table | |
156 | # define LIMn ecx // limit pointer | |
157 | # define LIM rcx // ... as integer offset from base | |
158 | ||
159 | # define NR ecx // number of rounds | |
160 | # define LRK eax // distance to last key | |
161 | # define LRKo rax // ... as address offset | |
162 | # define BLKOFF edx // block size in bytes | |
163 | # define BLKOFFo rdx // ... as address offset | |
164 | ||
165 | // We'll need the index registers, which belong to the caller in this | |
166 | // ABI. | |
167 | push rsi | |
168 | push rdi | |
169 | ||
170 | // Move arguments to more useful places. | |
171 | mov SI, r8 // key material | |
172 | mov CTX, rcx // context base pointer | |
173 | #endif | |
174 | ||
1a0c09c4 MW |
175 | // The initial round key material is taken directly from the input |
176 | // key, so copy it over. | |
0f23f75f MW |
177 | #if CPUFAM_AMD64 && ABI_SYSV |
178 | // We've been lucky. We already have a copy of the context pointer | |
179 | // in rdi, and the key size in ecx. | |
180 | add DI, w | |
181 | #else | |
182 | lea DI, [CTX + w] | |
183 | mov ecx, KSZ | |
184 | #endif | |
1a0c09c4 MW |
185 | rep movsd |
186 | ||
187 | // Find out other useful things. | |
0f23f75f MW |
188 | mov NKW, [CTX + nr] // number of rounds |
189 | add NKW, 1 | |
190 | imul NKW, BLKSZ // total key size in words | |
191 | #if !NKW_NEEDS_REFRESH | |
192 | // If we can't keep NKW for later, then we use the same register for | |
193 | // it and LIM, so this move is unnecessary. | |
194 | mov LIMn, NKW | |
195 | #endif | |
196 | sub LIMn, KSZ // offset by the key size | |
1a0c09c4 MW |
197 | |
198 | // Find the round constants. | |
199 | ldgot ecx | |
0f23f75f | 200 | leaext RCON, rijndael_rcon, ecx |
1a0c09c4 MW |
201 | |
202 | // Prepare for the main loop. | |
0f23f75f MW |
203 | lea SI, [CTX + w] |
204 | mov eax, [SI + 4*KSZo - 4] // most recent key word | |
205 | lea LIM, [SI + 4*LIM] // limit, offset by one key expansion | |
1a0c09c4 MW |
206 | |
207 | // Main key expansion loop. The first word of each key-length chunk | |
208 | // needs special treatment. | |
209 | // | |
210 | // This is rather tedious because the Intel `AESKEYGENASSIST' | |
211 | // instruction is very strangely shaped. Firstly, it wants to | |
212 | // operate on vast SSE registers, even though we're data-blocked from | |
213 | // doing more than operation at a time unless we're doing two key | |
214 | // schedules simultaneously -- and even then we can't do more than | |
215 | // two, because the instruction ignores two of its input words | |
216 | // entirely, and produces two different outputs for each of the other | |
217 | // two. And secondly it insists on taking the magic round constant | |
218 | // as an immediate, so it's kind of annoying if you're not | |
219 | // open-coding the whole thing. It's much easier to leave that as | |
220 | // zero and XOR in the round constant by hand. | |
221 | 9: movd xmm0, eax | |
47103664 | 222 | pshufd xmm0, xmm0, ROTR |
1a0c09c4 | 223 | aeskeygenassist xmm1, xmm0, 0 |
47103664 | 224 | pshufd xmm1, xmm1, ROTL |
1a0c09c4 | 225 | movd eax, xmm1 |
0f23f75f MW |
226 | xor eax, [SI] |
227 | xor al, [RCON] | |
228 | inc RCON | |
229 | mov [SI + 4*KSZo], eax | |
230 | add SI, 4 | |
231 | cmp SI, LIM | |
1a0c09c4 MW |
232 | jae 8f |
233 | ||
234 | // The next three words are simple... | |
0f23f75f MW |
235 | xor eax, [SI] |
236 | mov [SI + 4*KSZo], eax | |
237 | add SI, 4 | |
238 | cmp SI, LIM | |
1a0c09c4 MW |
239 | jae 8f |
240 | ||
241 | // (Word 2...) | |
0f23f75f MW |
242 | xor eax, [SI] |
243 | mov [SI + 4*KSZo], eax | |
244 | add SI, 4 | |
245 | cmp SI, LIM | |
1a0c09c4 MW |
246 | jae 8f |
247 | ||
248 | // (Word 3...) | |
0f23f75f MW |
249 | xor eax, [SI] |
250 | mov [SI + 4*KSZo], eax | |
251 | add SI, 4 | |
252 | cmp SI, LIM | |
1a0c09c4 MW |
253 | jae 8f |
254 | ||
255 | // Word 4. If the key is /more/ than 6 words long, then we must | |
256 | // apply a substitution here. | |
0f23f75f | 257 | cmp KSZ, 5 |
1a0c09c4 | 258 | jb 9b |
0f23f75f | 259 | cmp KSZ, 7 |
1a0c09c4 MW |
260 | jb 0f |
261 | movd xmm0, eax | |
47103664 | 262 | pshufd xmm0, xmm0, ROTL |
1a0c09c4 MW |
263 | aeskeygenassist xmm1, xmm0, 0 |
264 | movd eax, xmm1 | |
0f23f75f MW |
265 | 0: xor eax, [SI] |
266 | mov [SI + 4*KSZo], eax | |
267 | add SI, 4 | |
268 | cmp SI, LIM | |
1a0c09c4 MW |
269 | jae 8f |
270 | ||
271 | // (Word 5...) | |
0f23f75f | 272 | cmp KSZ, 6 |
1a0c09c4 | 273 | jb 9b |
0f23f75f MW |
274 | xor eax, [SI] |
275 | mov [SI + 4*KSZo], eax | |
276 | add SI, 4 | |
277 | cmp SI, LIM | |
1a0c09c4 MW |
278 | jae 8f |
279 | ||
280 | // (Word 6...) | |
0f23f75f | 281 | cmp KSZ, 7 |
1a0c09c4 | 282 | jb 9b |
0f23f75f MW |
283 | xor eax, [SI] |
284 | mov [SI + 4*KSZo], eax | |
285 | add SI, 4 | |
286 | cmp SI, LIM | |
1a0c09c4 MW |
287 | jae 8f |
288 | ||
289 | // (Word 7...) | |
0f23f75f | 290 | cmp KSZ, 8 |
1a0c09c4 | 291 | jb 9b |
0f23f75f MW |
292 | xor eax, [SI] |
293 | mov [SI + 4*KSZo], eax | |
294 | add SI, 4 | |
295 | cmp SI, LIM | |
1a0c09c4 MW |
296 | jae 8f |
297 | ||
298 | // Must be done by now. | |
299 | jmp 9b | |
300 | ||
301 | // Next job is to construct the decryption keys. The keys for the | |
302 | // first and last rounds don't need to be mangled, but the remaining | |
303 | // ones do -- and they all need to be reordered too. | |
304 | // | |
305 | // The plan of action, then, is to copy the final encryption round's | |
306 | // keys into place first, then to do each of the intermediate rounds | |
307 | // in reverse order, and finally do the first round. | |
308 | // | |
309 | // Do all of the heavy lifting with SSE registers. The order we're | |
310 | // doing this in means that it's OK if we read or write too much, and | |
311 | // there's easily enough buffer space for the over-enthusiastic reads | |
312 | // and writes because the context has space for 32-byte blocks, which | |
313 | // is our maximum and an exact fit for two SSE registers. | |
0f23f75f MW |
314 | 8: mov NR, [CTX + nr] // number of rounds |
315 | #if NKW_NEEDS_REFRESH | |
316 | mov BLKOFF, BLKSZ | |
317 | mov LRK, NR | |
318 | imul LRK, BLKOFF | |
319 | #else | |
320 | // If we retain NKW, then BLKSZ and BLKOFF are the same register | |
321 | // because we won't need the former again. | |
322 | mov LRK, NKW | |
323 | sub LRK, BLKSZ | |
324 | #endif | |
325 | lea DI, [CTX + wi] | |
326 | lea SI, [CTX + w + 4*LRKo] // last round's keys | |
327 | shl BLKOFF, 2 // block size (in bytes now) | |
1a0c09c4 MW |
328 | |
329 | // Copy the last encryption round's keys. | |
0f23f75f MW |
330 | movdqu xmm0, [SI] |
331 | movdqu [DI], xmm0 | |
332 | cmp BLKOFF, 16 | |
1a0c09c4 | 333 | jbe 9f |
0f23f75f MW |
334 | movdqu xmm0, [SI + 16] |
335 | movdqu [DI + 16], xmm0 | |
1a0c09c4 MW |
336 | |
337 | // Update the loop variables and stop if we've finished. | |
0f23f75f MW |
338 | 9: add DI, BLKOFFo |
339 | sub SI, BLKOFFo | |
340 | sub NR, 1 | |
1a0c09c4 MW |
341 | jbe 0f |
342 | ||
343 | // Do another middle round's keys... | |
0f23f75f | 344 | movdqu xmm0, [SI] |
1a0c09c4 | 345 | aesimc xmm0, xmm0 |
0f23f75f MW |
346 | movdqu [DI], xmm0 |
347 | cmp BLKOFF, 16 | |
1a0c09c4 | 348 | jbe 9b |
0f23f75f | 349 | movdqu xmm0, [SI + 16] |
1a0c09c4 | 350 | aesimc xmm0, xmm0 |
0f23f75f | 351 | movdqu [DI + 16], xmm0 |
1a0c09c4 MW |
352 | jmp 9b |
353 | ||
354 | // Finally do the first encryption round. | |
0f23f75f MW |
355 | 0: movdqu xmm0, [SI] |
356 | movdqu [DI], xmm0 | |
357 | cmp BLKOFF, 16 | |
1a0c09c4 | 358 | jbe 0f |
0f23f75f MW |
359 | movdqu xmm0, [SI + 16] |
360 | movdqu [DI + 16], xmm0 | |
1a0c09c4 MW |
361 | |
362 | // If the block size is not exactly four words then we must end-swap | |
363 | // everything. We can use fancy SSE toys for this. | |
0f23f75f | 364 | 0: cmp BLKOFF, 16 |
1a0c09c4 MW |
365 | je 0f |
366 | ||
367 | // Find the byte-reordering table. | |
368 | ldgot ecx | |
8d6ca554 | 369 | movdqa xmm5, [INTADDR(endswap_tab, ecx)] |
1a0c09c4 | 370 | |
0f23f75f | 371 | #if NKW_NEEDS_REFRESH |
1a0c09c4 MW |
372 | // Calculate the number of subkey words again. (It's a good job |
373 | // we've got a fast multiplier.) | |
0f23f75f MW |
374 | mov NKW, [CTX + nr] |
375 | add NKW, 1 | |
376 | imul NKW, BLKSZ | |
377 | #endif | |
1a0c09c4 MW |
378 | |
379 | // End-swap the encryption keys. | |
0f23f75f MW |
380 | mov ecx, NKW |
381 | lea SI, [CTX + w] | |
1a0c09c4 MW |
382 | call endswap_block |
383 | ||
384 | // And the decryption keys. | |
0f23f75f MW |
385 | mov ecx, NKW |
386 | lea SI, [CTX + wi] | |
1a0c09c4 MW |
387 | call endswap_block |
388 | ||
0f23f75f MW |
389 | 0: // All done. |
390 | #if CPUFAM_X86 | |
391 | pop edi | |
1a0c09c4 MW |
392 | pop esi |
393 | pop ebx | |
394 | pop ebp | |
0f23f75f MW |
395 | #endif |
396 | #if CPUFAM_AMD64 && ABI_WIN | |
397 | pop rdi | |
398 | pop rsi | |
399 | #endif | |
1a0c09c4 MW |
400 | ret |
401 | ||
402 | .align 16 | |
403 | endswap_block: | |
0f23f75f | 404 | // End-swap ECX words starting at SI. The end-swapping table is |
8d6ca554 | 405 | // already loaded into XMM5; and it's OK to work in 16-byte chunks. |
0f23f75f | 406 | movdqu xmm1, [SI] |
8d6ca554 | 407 | pshufb xmm1, xmm5 |
0f23f75f MW |
408 | movdqu [SI], xmm1 |
409 | add SI, 16 | |
1a0c09c4 MW |
410 | sub ecx, 4 |
411 | ja endswap_block | |
412 | ret | |
413 | ||
0f23f75f MW |
414 | #undef CTX |
415 | #undef BLKSZ | |
416 | #undef SI | |
417 | #undef DI | |
418 | #undef KSZ | |
419 | #undef KSZo | |
420 | #undef RCON | |
421 | #undef LIMn | |
422 | #undef LIM | |
423 | #undef NR | |
424 | #undef LRK | |
425 | #undef LRKo | |
426 | #undef BLKOFF | |
427 | #undef BLKOFFo | |
428 | ||
1a0c09c4 MW |
429 | ENDFUNC |
430 | ||
431 | ///-------------------------------------------------------------------------- | |
432 | /// Encrypting and decrypting blocks. | |
433 | ||
e297526c | 434 | .macro encdec op, aes, koff |
0f23f75f | 435 | FUNC(rijndael_\op\()_x86ish_aesni) |
1a0c09c4 MW |
436 | |
437 | // Find the magic endianness-swapping table. | |
438 | ldgot ecx | |
8d6ca554 | 439 | movdqa xmm5, [INTADDR(endswap_tab, ecx)] |
1a0c09c4 | 440 | |
0f23f75f MW |
441 | #if CPUFAM_X86 |
442 | // Arguments come in on the stack, and need to be collected. We | |
443 | // don't have a shortage of registers. | |
444 | ||
445 | # define K ecx | |
446 | # define SRC edx | |
447 | # define DST edx | |
448 | # define NR eax | |
449 | ||
450 | mov K, [esp + 4] | |
451 | mov SRC, [esp + 8] | |
452 | #endif | |
453 | ||
454 | #if CPUFAM_AMD64 && ABI_SYSV | |
455 | // Arguments come in registers. All is good. | |
456 | ||
457 | # define K rdi | |
458 | # define SRC rsi | |
459 | # define DST rdx | |
460 | # define NR eax | |
461 | #endif | |
462 | ||
463 | #if CPUFAM_AMD64 && ABI_WIN | |
464 | // Arguments come in different registers. | |
465 | ||
466 | # define K rcx | |
467 | # define SRC rdx | |
468 | # define DST r8 | |
469 | # define NR eax | |
470 | #endif | |
471 | ||
472 | // Initial setup. | |
473 | movdqu xmm0, [SRC] | |
8d6ca554 | 474 | pshufb xmm0, xmm5 |
0f23f75f MW |
475 | mov NR, [K + nr] |
476 | add K, \koff | |
1a0c09c4 MW |
477 | |
478 | // Initial whitening. | |
0f23f75f MW |
479 | movdqu xmm1, [K] |
480 | add K, 16 | |
1a0c09c4 MW |
481 | pxor xmm0, xmm1 |
482 | ||
483 | // Dispatch to the correct code. | |
0f23f75f | 484 | cmp NR, 10 |
e297526c | 485 | je 10f |
1a0c09c4 | 486 | jb bogus |
0f23f75f | 487 | cmp NR, 14 |
e297526c | 488 | je 14f |
1a0c09c4 | 489 | ja bogus |
0f23f75f | 490 | cmp NR, 12 |
e297526c MW |
491 | je 12f |
492 | jb 11f | |
493 | jmp 13f | |
1a0c09c4 MW |
494 | |
495 | .align 2 | |
496 | ||
497 | // 14 rounds... | |
0f23f75f MW |
498 | 14: movdqu xmm1, [K] |
499 | add K, 16 | |
e297526c | 500 | \aes xmm0, xmm1 |
1a0c09c4 MW |
501 | |
502 | // 13 rounds... | |
0f23f75f MW |
503 | 13: movdqu xmm1, [K] |
504 | add K, 16 | |
e297526c | 505 | \aes xmm0, xmm1 |
1a0c09c4 MW |
506 | |
507 | // 12 rounds... | |
0f23f75f MW |
508 | 12: movdqu xmm1, [K] |
509 | add K, 16 | |
e297526c | 510 | \aes xmm0, xmm1 |
1a0c09c4 MW |
511 | |
512 | // 11 rounds... | |
0f23f75f MW |
513 | 11: movdqu xmm1, [K] |
514 | add K, 16 | |
e297526c | 515 | \aes xmm0, xmm1 |
1a0c09c4 MW |
516 | |
517 | // 10 rounds... | |
0f23f75f | 518 | 10: movdqu xmm1, [K] |
e297526c | 519 | \aes xmm0, xmm1 |
1a0c09c4 MW |
520 | |
521 | // 9 rounds... | |
0f23f75f | 522 | movdqu xmm1, [K + 16] |
e297526c | 523 | \aes xmm0, xmm1 |
1a0c09c4 MW |
524 | |
525 | // 8 rounds... | |
0f23f75f | 526 | movdqu xmm1, [K + 32] |
e297526c | 527 | \aes xmm0, xmm1 |
1a0c09c4 MW |
528 | |
529 | // 7 rounds... | |
0f23f75f | 530 | movdqu xmm1, [K + 48] |
e297526c | 531 | \aes xmm0, xmm1 |
1a0c09c4 MW |
532 | |
533 | // 6 rounds... | |
0f23f75f | 534 | movdqu xmm1, [K + 64] |
e297526c | 535 | \aes xmm0, xmm1 |
1a0c09c4 MW |
536 | |
537 | // 5 rounds... | |
0f23f75f | 538 | movdqu xmm1, [K + 80] |
e297526c | 539 | \aes xmm0, xmm1 |
1a0c09c4 MW |
540 | |
541 | // 4 rounds... | |
0f23f75f | 542 | movdqu xmm1, [K + 96] |
e297526c | 543 | \aes xmm0, xmm1 |
1a0c09c4 MW |
544 | |
545 | // 3 rounds... | |
0f23f75f | 546 | movdqu xmm1, [K + 112] |
e297526c | 547 | \aes xmm0, xmm1 |
1a0c09c4 MW |
548 | |
549 | // 2 rounds... | |
0f23f75f | 550 | movdqu xmm1, [K + 128] |
e297526c | 551 | \aes xmm0, xmm1 |
1a0c09c4 MW |
552 | |
553 | // Final round... | |
0f23f75f | 554 | movdqu xmm1, [K + 144] |
e297526c | 555 | \aes\()last xmm0, xmm1 |
1a0c09c4 MW |
556 | |
557 | // Unpermute the ciphertext block and store it. | |
8d6ca554 | 558 | pshufb xmm0, xmm5 |
0f23f75f MW |
559 | #if CPUFAM_X86 |
560 | mov DST, [esp + 12] | |
561 | #endif | |
562 | movdqu [DST], xmm0 | |
1a0c09c4 MW |
563 | |
564 | // And we're done. | |
565 | ret | |
566 | ||
0f23f75f MW |
567 | #undef K |
568 | #undef SRC | |
569 | #undef DST | |
570 | #undef NR | |
571 | ||
1a0c09c4 | 572 | ENDFUNC |
e297526c | 573 | .endm |
1a0c09c4 | 574 | |
e297526c MW |
575 | encdec eblk, aesenc, w |
576 | encdec dblk, aesdec, wi | |
1a0c09c4 MW |
577 | |
578 | ///-------------------------------------------------------------------------- | |
579 | /// Random utilities. | |
580 | ||
581 | .align 16 | |
582 | // Abort the process because of a programming error. Indirecting | |
583 | // through this point serves several purposes: (a) by CALLing, rather | |
584 | // than branching to, `abort', we can save the return address, which | |
585 | // might at least provide a hint as to what went wrong; (b) we don't | |
586 | // have conditional CALLs (and they'd be big anyway); and (c) we can | |
587 | // write a HLT here as a backstop against `abort' being mad. | |
588 | bogus: callext F(abort) | |
589 | 0: hlt | |
590 | jmp 0b | |
591 | ||
592 | gotaux ecx | |
593 | ||
594 | ///-------------------------------------------------------------------------- | |
595 | /// Data tables. | |
596 | ||
597 | .align 16 | |
598 | endswap_tab: | |
599 | .byte 3, 2, 1, 0 | |
600 | .byte 7, 6, 5, 4 | |
601 | .byte 11, 10, 9, 8 | |
602 | .byte 15, 14, 13, 12 | |
603 | ||
604 | ///----- That's all, folks -------------------------------------------------- |