1/*
2 * Copyright (c) 2016 Thomas Pornin <pornin@bolet.org>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining
5 * a copy of this software and associated documentation files (the
6 * "Software"), to deal in the Software without restriction, including
7 * without limitation the rights to use, copy, modify, merge, publish,
8 * distribute, sublicense, and/or sell copies of the Software, and to
9 * permit persons to whom the Software is furnished to do so, subject to
10 * the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
24
25#include <sys/cdefs.h>
26__KERNEL_RCSID(1, "$NetBSD: aes_sse2.c,v 1.2 2020/06/30 20:32:11 riastradh Exp $");
27
28#include <sys/types.h>
29
30#ifdef _KERNEL
31#include <lib/libkern/libkern.h>
32#else
33#include <stdint.h>
34#include <string.h>
35#endif
36
37#include "aes_sse2_impl.h"
38
39static void
40br_range_dec32le(uint32_t *p32, size_t nwords, const void *v)
41{
42	const uint8_t *p8 = v;
43
44	while (nwords --> 0) {
45		uint32_t x0 = *p8++;
46		uint32_t x1 = *p8++;
47		uint32_t x2 = *p8++;
48		uint32_t x3 = *p8++;
49
50		*p32++ = x0 | (x1 << 8) | (x2 << 16) | (x3 << 24);
51	}
52}
53
54void
55aes_sse2_bitslice_Sbox(__m128i q[static 4])
56{
57	__m128i x0, x1, x2, x3, x4, x5, x6, x7;
58	__m128i y1, y2, y3, y4, y5, y6, y7, y8, y9;
59	__m128i y10, y11, y12, y13, y14, y15, y16, y17, y18, y19;
60	__m128i y20, y21;
61	__m128i z0, z1, z2, z3, z4, z5, z6, z7, z8, z9;
62	__m128i z10, z11, z12, z13, z14, z15, z16, z17;
63	__m128i t0, t1, t2, t3, t4, t5, t6, t7, t8, t9;
64	__m128i t10, t11, t12, t13, t14, t15, t16, t17, t18, t19;
65	__m128i t20, t21, t22, t23, t24, t25, t26, t27, t28, t29;
66	__m128i t30, t31, t32, t33, t34, t35, t36, t37, t38, t39;
67	__m128i t40, t41, t42, t43, t44, t45, t46, t47, t48, t49;
68	__m128i t50, t51, t52, t53, t54, t55, t56, t57, t58, t59;
69	__m128i t60, t61, t62, t63, t64, t65, t66, t67;
70	__m128i s0, s1, s2, s3, s4, s5, s6, s7;
71
72	x0 = _mm_shuffle_epi32(q[3], 0x0e);
73	x1 = _mm_shuffle_epi32(q[2], 0x0e);
74	x2 = _mm_shuffle_epi32(q[1], 0x0e);
75	x3 = _mm_shuffle_epi32(q[0], 0x0e);
76	x4 = q[3];
77	x5 = q[2];
78	x6 = q[1];
79	x7 = q[0];
80
81	/*
82	 * Top linear transformation.
83	 */
84	y14 = x3 ^ x5;
85	y13 = x0 ^ x6;
86	y9 = x0 ^ x3;
87	y8 = x0 ^ x5;
88	t0 = x1 ^ x2;
89	y1 = t0 ^ x7;
90	y4 = y1 ^ x3;
91	y12 = y13 ^ y14;
92	y2 = y1 ^ x0;
93	y5 = y1 ^ x6;
94	y3 = y5 ^ y8;
95	t1 = x4 ^ y12;
96	y15 = t1 ^ x5;
97	y20 = t1 ^ x1;
98	y6 = y15 ^ x7;
99	y10 = y15 ^ t0;
100	y11 = y20 ^ y9;
101	y7 = x7 ^ y11;
102	y17 = y10 ^ y11;
103	y19 = y10 ^ y8;
104	y16 = t0 ^ y11;
105	y21 = y13 ^ y16;
106	y18 = x0 ^ y16;
107
108	/*
109	 * Non-linear section.
110	 */
111	t2 = y12 & y15;
112	t3 = y3 & y6;
113	t4 = t3 ^ t2;
114	t5 = y4 & x7;
115	t6 = t5 ^ t2;
116	t7 = y13 & y16;
117	t8 = y5 & y1;
118	t9 = t8 ^ t7;
119	t10 = y2 & y7;
120	t11 = t10 ^ t7;
121	t12 = y9 & y11;
122	t13 = y14 & y17;
123	t14 = t13 ^ t12;
124	t15 = y8 & y10;
125	t16 = t15 ^ t12;
126	t17 = t4 ^ t14;
127	t18 = t6 ^ t16;
128	t19 = t9 ^ t14;
129	t20 = t11 ^ t16;
130	t21 = t17 ^ y20;
131	t22 = t18 ^ y19;
132	t23 = t19 ^ y21;
133	t24 = t20 ^ y18;
134
135	t25 = t21 ^ t22;
136	t26 = t21 & t23;
137	t27 = t24 ^ t26;
138	t28 = t25 & t27;
139	t29 = t28 ^ t22;
140	t30 = t23 ^ t24;
141	t31 = t22 ^ t26;
142	t32 = t31 & t30;
143	t33 = t32 ^ t24;
144	t34 = t23 ^ t33;
145	t35 = t27 ^ t33;
146	t36 = t24 & t35;
147	t37 = t36 ^ t34;
148	t38 = t27 ^ t36;
149	t39 = t29 & t38;
150	t40 = t25 ^ t39;
151
152	t41 = t40 ^ t37;
153	t42 = t29 ^ t33;
154	t43 = t29 ^ t40;
155	t44 = t33 ^ t37;
156	t45 = t42 ^ t41;
157	z0 = t44 & y15;
158	z1 = t37 & y6;
159	z2 = t33 & x7;
160	z3 = t43 & y16;
161	z4 = t40 & y1;
162	z5 = t29 & y7;
163	z6 = t42 & y11;
164	z7 = t45 & y17;
165	z8 = t41 & y10;
166	z9 = t44 & y12;
167	z10 = t37 & y3;
168	z11 = t33 & y4;
169	z12 = t43 & y13;
170	z13 = t40 & y5;
171	z14 = t29 & y2;
172	z15 = t42 & y9;
173	z16 = t45 & y14;
174	z17 = t41 & y8;
175
176	/*
177	 * Bottom linear transformation.
178	 */
179	t46 = z15 ^ z16;
180	t47 = z10 ^ z11;
181	t48 = z5 ^ z13;
182	t49 = z9 ^ z10;
183	t50 = z2 ^ z12;
184	t51 = z2 ^ z5;
185	t52 = z7 ^ z8;
186	t53 = z0 ^ z3;
187	t54 = z6 ^ z7;
188	t55 = z16 ^ z17;
189	t56 = z12 ^ t48;
190	t57 = t50 ^ t53;
191	t58 = z4 ^ t46;
192	t59 = z3 ^ t54;
193	t60 = t46 ^ t57;
194	t61 = z14 ^ t57;
195	t62 = t52 ^ t58;
196	t63 = t49 ^ t58;
197	t64 = z4 ^ t59;
198	t65 = t61 ^ t62;
199	t66 = z1 ^ t63;
200	s0 = t59 ^ t63;
201	s6 = t56 ^ ~t62;
202	s7 = t48 ^ ~t60;
203	t67 = t64 ^ t65;
204	s3 = t53 ^ t66;
205	s4 = t51 ^ t66;
206	s5 = t47 ^ t65;
207	s1 = t64 ^ ~s3;
208	s2 = t55 ^ ~t67;
209
210	q[3] = _mm_unpacklo_epi64(s4, s0);
211	q[2] = _mm_unpacklo_epi64(s5, s1);
212	q[1] = _mm_unpacklo_epi64(s6, s2);
213	q[0] = _mm_unpacklo_epi64(s7, s3);
214}
215
216void
217aes_sse2_ortho(__m128i q[static 4])
218{
219#define SWAPN(cl, ch, s, x, y)   do { \
220		__m128i a, b; \
221		a = (x); \
222		b = (y); \
223		(x) = (a & _mm_set1_epi64x(cl)) | \
224		    _mm_slli_epi64(b & _mm_set1_epi64x(cl), (s)); \
225		(y) = _mm_srli_epi64(a & _mm_set1_epi64x(ch), (s)) | \
226		    (b & _mm_set1_epi64x(ch)); \
227	} while (0)
228
229#define SWAP2(x, y)    SWAPN(0x5555555555555555, 0xAAAAAAAAAAAAAAAA,  1, x, y)
230#define SWAP4(x, y)    SWAPN(0x3333333333333333, 0xCCCCCCCCCCCCCCCC,  2, x, y)
231#define SWAP8(x, y)    SWAPN(0x0F0F0F0F0F0F0F0F, 0xF0F0F0F0F0F0F0F0,  4, x, y)
232
233	SWAP2(q[0], q[1]);
234	SWAP2(q[2], q[3]);
235
236	SWAP4(q[0], q[2]);
237	SWAP4(q[1], q[3]);
238
239	__m128i q0 = q[0];
240	__m128i q1 = q[1];
241	__m128i q2 = q[2];
242	__m128i q3 = q[3];
243	__m128i q4 = _mm_shuffle_epi32(q[0], 0x0e);
244	__m128i q5 = _mm_shuffle_epi32(q[1], 0x0e);
245	__m128i q6 = _mm_shuffle_epi32(q[2], 0x0e);
246	__m128i q7 = _mm_shuffle_epi32(q[3], 0x0e);
247	SWAP8(q0, q4);
248	SWAP8(q1, q5);
249	SWAP8(q2, q6);
250	SWAP8(q3, q7);
251	q[0] = _mm_unpacklo_epi64(q0, q4);
252	q[1] = _mm_unpacklo_epi64(q1, q5);
253	q[2] = _mm_unpacklo_epi64(q2, q6);
254	q[3] = _mm_unpacklo_epi64(q3, q7);
255}
256
257__m128i
258aes_sse2_interleave_in(__m128i w)
259{
260	__m128i lo, hi;
261
262	lo = _mm_shuffle_epi32(w, 0x10);
263	hi = _mm_shuffle_epi32(w, 0x32);
264	lo &= _mm_set1_epi64x(0x00000000FFFFFFFF);
265	hi &= _mm_set1_epi64x(0x00000000FFFFFFFF);
266	lo |= _mm_slli_epi64(lo, 16);
267	hi |= _mm_slli_epi64(hi, 16);
268	lo &= _mm_set1_epi32(0x0000FFFF);
269	hi &= _mm_set1_epi32(0x0000FFFF);
270	lo |= _mm_slli_epi64(lo, 8);
271	hi |= _mm_slli_epi64(hi, 8);
272	lo &= _mm_set1_epi16(0x00FF);
273	hi &= _mm_set1_epi16(0x00FF);
274	return lo | _mm_slli_epi64(hi, 8);
275}
276
277__m128i
278aes_sse2_interleave_out(__m128i q)
279{
280	__m128i lo, hi;
281
282	lo = q;
283	hi = _mm_srli_si128(q, 1);
284	lo &= _mm_set1_epi16(0x00FF);
285	hi &= _mm_set1_epi16(0x00FF);
286	lo |= _mm_srli_epi64(lo, 8);
287	hi |= _mm_srli_epi64(hi, 8);
288	lo &= _mm_set1_epi32(0x0000FFFF);
289	hi &= _mm_set1_epi32(0x0000FFFF);
290	lo |= _mm_srli_epi64(lo, 16);
291	hi |= _mm_srli_epi64(hi, 16);
292	return (__m128i)_mm_shuffle_ps((__m128)lo, (__m128)hi, 0x88);
293}
294
295static const unsigned char Rcon[] = {
296	0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1B, 0x36
297};
298
299static uint32_t
300sub_word(uint32_t x)
301{
302	__m128i q[4];
303	uint32_t y;
304
305	memset(q, 0, sizeof(q));
306	q[0] = _mm_loadu_si32(&x);
307	aes_sse2_ortho(q);
308	aes_sse2_bitslice_Sbox(q);
309	aes_sse2_ortho(q);
310	_mm_storeu_si32(&y, q[0]);
311	return y;
312}
313
314unsigned
315aes_sse2_keysched(uint64_t *comp_skey, const void *key, size_t key_len)
316{
317	unsigned num_rounds;
318	int i, j, k, nk, nkf;
319	uint32_t tmp;
320	uint32_t skey[60];
321
322	switch (key_len) {
323	case 16:
324		num_rounds = 10;
325		break;
326	case 24:
327		num_rounds = 12;
328		break;
329	case 32:
330		num_rounds = 14;
331		break;
332	default:
333		/* abort(); */
334		return 0;
335	}
336	nk = (int)(key_len >> 2);
337	nkf = (int)((num_rounds + 1) << 2);
338	br_range_dec32le(skey, (key_len >> 2), key);
339	tmp = skey[(key_len >> 2) - 1];
340	for (i = nk, j = 0, k = 0; i < nkf; i ++) {
341		if (j == 0) {
342			tmp = (tmp << 24) | (tmp >> 8);
343			tmp = sub_word(tmp) ^ Rcon[k];
344		} else if (nk > 6 && j == 4) {
345			tmp = sub_word(tmp);
346		}
347		tmp ^= skey[i - nk];
348		skey[i] = tmp;
349		if (++ j == nk) {
350			j = 0;
351			k ++;
352		}
353	}
354
355	for (i = 0, j = 0; i < nkf; i += 4, j += 2) {
356		__m128i q[4], q0, q1, q2, q3, q4, q5, q6, q7;
357		__m128i w;
358
359		w = _mm_loadu_epi8(skey + i);
360		q[0] = q[1] = q[2] = q[3] = aes_sse2_interleave_in(w);
361		aes_sse2_ortho(q);
362		q0 = q[0] & _mm_set1_epi64x(0x1111111111111111);
363		q1 = q[1] & _mm_set1_epi64x(0x2222222222222222);
364		q2 = q[2] & _mm_set1_epi64x(0x4444444444444444);
365		q3 = q[3] & _mm_set1_epi64x(0x8888888888888888);
366		q4 = _mm_shuffle_epi32(q0, 0x0e);
367		q5 = _mm_shuffle_epi32(q1, 0x0e);
368		q6 = _mm_shuffle_epi32(q2, 0x0e);
369		q7 = _mm_shuffle_epi32(q3, 0x0e);
370		_mm_storeu_si64(&comp_skey[j + 0], q0 | q1 | q2 | q3);
371		_mm_storeu_si64(&comp_skey[j + 1], q4 | q5 | q6 | q7);
372	}
373	return num_rounds;
374}
375
376void
377aes_sse2_skey_expand(uint64_t *skey,
378	unsigned num_rounds, const uint64_t *comp_skey)
379{
380	unsigned u, v, n;
381
382	n = (num_rounds + 1) << 1;
383	for (u = 0, v = 0; u < n; u ++, v += 4) {
384		__m128i x0, x1, x2, x3;
385
386		x0 = x1 = x2 = x3 = _mm_loadu_si64(&comp_skey[u]);
387		x0 &= 0x1111111111111111;
388		x1 &= 0x2222222222222222;
389		x2 &= 0x4444444444444444;
390		x3 &= 0x8888888888888888;
391		x1 = _mm_srli_epi64(x1, 1);
392		x2 = _mm_srli_epi64(x2, 2);
393		x3 = _mm_srli_epi64(x3, 3);
394		x0 = _mm_sub_epi64(_mm_slli_epi64(x0, 4), x0);
395		x1 = _mm_sub_epi64(_mm_slli_epi64(x1, 4), x1);
396		x2 = _mm_sub_epi64(_mm_slli_epi64(x2, 4), x2);
397		x3 = _mm_sub_epi64(_mm_slli_epi64(x3, 4), x3);
398		_mm_storeu_si64(&skey[v + 0], x0);
399		_mm_storeu_si64(&skey[v + 1], x1);
400		_mm_storeu_si64(&skey[v + 2], x2);
401		_mm_storeu_si64(&skey[v + 3], x3);
402	}
403}
404