1/*
2 * x86_64/AVX/AES-NI assembler implementation of Camellia
3 *
4 * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 */
12
13/*
14 * Version licensed under 2-clause BSD License is available at:
15 *	http://koti.mbnet.fi/axh/crypto/camellia-BSD-1.2.0-aesni1.tar.xz
16 */
17
18#include <linux/linkage.h>
19#include <asm/frame.h>
20
21#define CAMELLIA_TABLE_BYTE_LEN 272
22
23/* struct camellia_ctx: */
24#define key_table 0
25#define key_length CAMELLIA_TABLE_BYTE_LEN
26
27/* register macros */
28#define CTX %rdi
29
30/**********************************************************************
31  16-way camellia
32 **********************************************************************/
33#define filter_8bit(x, lo_t, hi_t, mask4bit, tmp0) \
34	vpand x, mask4bit, tmp0; \
35	vpandn x, mask4bit, x; \
36	vpsrld $4, x, x; \
37	\
38	vpshufb tmp0, lo_t, tmp0; \
39	vpshufb x, hi_t, x; \
40	vpxor tmp0, x, x;
41
42/*
43 * IN:
44 *   x0..x7: byte-sliced AB state
45 *   mem_cd: register pointer storing CD state
46 *   key: index for key material
47 * OUT:
48 *   x0..x7: new byte-sliced CD state
49 */
50#define roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, t0, t1, t2, t3, t4, t5, t6, \
51		  t7, mem_cd, key) \
52	/* \
53	 * S-function with AES subbytes \
54	 */ \
55	vmovdqa .Linv_shift_row(%rip), t4; \
56	vbroadcastss .L0f0f0f0f(%rip), t7; \
57	vmovdqa .Lpre_tf_lo_s1(%rip), t0; \
58	vmovdqa .Lpre_tf_hi_s1(%rip), t1; \
59	\
60	/* AES inverse shift rows */ \
61	vpshufb t4, x0, x0; \
62	vpshufb t4, x7, x7; \
63	vpshufb t4, x1, x1; \
64	vpshufb t4, x4, x4; \
65	vpshufb t4, x2, x2; \
66	vpshufb t4, x5, x5; \
67	vpshufb t4, x3, x3; \
68	vpshufb t4, x6, x6; \
69	\
70	/* prefilter sboxes 1, 2 and 3 */ \
71	vmovdqa .Lpre_tf_lo_s4(%rip), t2; \
72	vmovdqa .Lpre_tf_hi_s4(%rip), t3; \
73	filter_8bit(x0, t0, t1, t7, t6); \
74	filter_8bit(x7, t0, t1, t7, t6); \
75	filter_8bit(x1, t0, t1, t7, t6); \
76	filter_8bit(x4, t0, t1, t7, t6); \
77	filter_8bit(x2, t0, t1, t7, t6); \
78	filter_8bit(x5, t0, t1, t7, t6); \
79	\
80	/* prefilter sbox 4 */ \
81	vpxor t4, t4, t4; \
82	filter_8bit(x3, t2, t3, t7, t6); \
83	filter_8bit(x6, t2, t3, t7, t6); \
84	\
85	/* AES subbytes + AES shift rows */ \
86	vmovdqa .Lpost_tf_lo_s1(%rip), t0; \
87	vmovdqa .Lpost_tf_hi_s1(%rip), t1; \
88	vaesenclast t4, x0, x0; \
89	vaesenclast t4, x7, x7; \
90	vaesenclast t4, x1, x1; \
91	vaesenclast t4, x4, x4; \
92	vaesenclast t4, x2, x2; \
93	vaesenclast t4, x5, x5; \
94	vaesenclast t4, x3, x3; \
95	vaesenclast t4, x6, x6; \
96	\
97	/* postfilter sboxes 1 and 4 */ \
98	vmovdqa .Lpost_tf_lo_s3(%rip), t2; \
99	vmovdqa .Lpost_tf_hi_s3(%rip), t3; \
100	filter_8bit(x0, t0, t1, t7, t6); \
101	filter_8bit(x7, t0, t1, t7, t6); \
102	filter_8bit(x3, t0, t1, t7, t6); \
103	filter_8bit(x6, t0, t1, t7, t6); \
104	\
105	/* postfilter sbox 3 */ \
106	vmovdqa .Lpost_tf_lo_s2(%rip), t4; \
107	vmovdqa .Lpost_tf_hi_s2(%rip), t5; \
108	filter_8bit(x2, t2, t3, t7, t6); \
109	filter_8bit(x5, t2, t3, t7, t6); \
110	\
111	vpxor t6, t6, t6; \
112	vmovq key, t0; \
113	\
114	/* postfilter sbox 2 */ \
115	filter_8bit(x1, t4, t5, t7, t2); \
116	filter_8bit(x4, t4, t5, t7, t2); \
117	\
118	vpsrldq $5, t0, t5; \
119	vpsrldq $1, t0, t1; \
120	vpsrldq $2, t0, t2; \
121	vpsrldq $3, t0, t3; \
122	vpsrldq $4, t0, t4; \
123	vpshufb t6, t0, t0; \
124	vpshufb t6, t1, t1; \
125	vpshufb t6, t2, t2; \
126	vpshufb t6, t3, t3; \
127	vpshufb t6, t4, t4; \
128	vpsrldq $2, t5, t7; \
129	vpshufb t6, t7, t7; \
130	\
131	/* \
132	 * P-function \
133	 */ \
134	vpxor x5, x0, x0; \
135	vpxor x6, x1, x1; \
136	vpxor x7, x2, x2; \
137	vpxor x4, x3, x3; \
138	\
139	vpxor x2, x4, x4; \
140	vpxor x3, x5, x5; \
141	vpxor x0, x6, x6; \
142	vpxor x1, x7, x7; \
143	\
144	vpxor x7, x0, x0; \
145	vpxor x4, x1, x1; \
146	vpxor x5, x2, x2; \
147	vpxor x6, x3, x3; \
148	\
149	vpxor x3, x4, x4; \
150	vpxor x0, x5, x5; \
151	vpxor x1, x6, x6; \
152	vpxor x2, x7, x7; /* note: high and low parts swapped */ \
153	\
154	/* \
155	 * Add key material and result to CD (x becomes new CD) \
156	 */ \
157	\
158	vpxor t3, x4, x4; \
159	vpxor 0 * 16(mem_cd), x4, x4; \
160	\
161	vpxor t2, x5, x5; \
162	vpxor 1 * 16(mem_cd), x5, x5; \
163	\
164	vpsrldq $1, t5, t3; \
165	vpshufb t6, t5, t5; \
166	vpshufb t6, t3, t6; \
167	\
168	vpxor t1, x6, x6; \
169	vpxor 2 * 16(mem_cd), x6, x6; \
170	\
171	vpxor t0, x7, x7; \
172	vpxor 3 * 16(mem_cd), x7, x7; \
173	\
174	vpxor t7, x0, x0; \
175	vpxor 4 * 16(mem_cd), x0, x0; \
176	\
177	vpxor t6, x1, x1; \
178	vpxor 5 * 16(mem_cd), x1, x1; \
179	\
180	vpxor t5, x2, x2; \
181	vpxor 6 * 16(mem_cd), x2, x2; \
182	\
183	vpxor t4, x3, x3; \
184	vpxor 7 * 16(mem_cd), x3, x3;
185
186/*
187 * Size optimization... with inlined roundsm16, binary would be over 5 times
188 * larger and would only be 0.5% faster (on sandy-bridge).
189 */
190.align 8
191SYM_FUNC_START_LOCAL(roundsm16_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
192	roundsm16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
193		  %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14, %xmm15,
194		  %rcx, (%r9));
195	RET;
196SYM_FUNC_END(roundsm16_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
197
198.align 8
199SYM_FUNC_START_LOCAL(roundsm16_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
200	roundsm16(%xmm4, %xmm5, %xmm6, %xmm7, %xmm0, %xmm1, %xmm2, %xmm3,
201		  %xmm12, %xmm13, %xmm14, %xmm15, %xmm8, %xmm9, %xmm10, %xmm11,
202		  %rax, (%r9));
203	RET;
204SYM_FUNC_END(roundsm16_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
205
206/*
207 * IN/OUT:
208 *  x0..x7: byte-sliced AB state preloaded
209 *  mem_ab: byte-sliced AB state in memory
210 *  mem_cb: byte-sliced CD state in memory
211 */
212#define two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
213		      y6, y7, mem_ab, mem_cd, i, dir, store_ab) \
214	leaq (key_table + (i) * 8)(CTX), %r9; \
215	call roundsm16_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd; \
216	\
217	vmovdqu x4, 0 * 16(mem_cd); \
218	vmovdqu x5, 1 * 16(mem_cd); \
219	vmovdqu x6, 2 * 16(mem_cd); \
220	vmovdqu x7, 3 * 16(mem_cd); \
221	vmovdqu x0, 4 * 16(mem_cd); \
222	vmovdqu x1, 5 * 16(mem_cd); \
223	vmovdqu x2, 6 * 16(mem_cd); \
224	vmovdqu x3, 7 * 16(mem_cd); \
225	\
226	leaq (key_table + ((i) + (dir)) * 8)(CTX), %r9; \
227	call roundsm16_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab; \
228	\
229	store_ab(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab);
230
231#define dummy_store(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) /* do nothing */
232
233#define store_ab_state(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) \
234	/* Store new AB state */ \
235	vmovdqu x0, 0 * 16(mem_ab); \
236	vmovdqu x1, 1 * 16(mem_ab); \
237	vmovdqu x2, 2 * 16(mem_ab); \
238	vmovdqu x3, 3 * 16(mem_ab); \
239	vmovdqu x4, 4 * 16(mem_ab); \
240	vmovdqu x5, 5 * 16(mem_ab); \
241	vmovdqu x6, 6 * 16(mem_ab); \
242	vmovdqu x7, 7 * 16(mem_ab);
243
244#define enc_rounds16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
245		      y6, y7, mem_ab, mem_cd, i) \
246	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
247		      y6, y7, mem_ab, mem_cd, (i) + 2, 1, store_ab_state); \
248	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
249		      y6, y7, mem_ab, mem_cd, (i) + 4, 1, store_ab_state); \
250	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
251		      y6, y7, mem_ab, mem_cd, (i) + 6, 1, dummy_store);
252
253#define dec_rounds16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
254		      y6, y7, mem_ab, mem_cd, i) \
255	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
256		      y6, y7, mem_ab, mem_cd, (i) + 7, -1, store_ab_state); \
257	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
258		      y6, y7, mem_ab, mem_cd, (i) + 5, -1, store_ab_state); \
259	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
260		      y6, y7, mem_ab, mem_cd, (i) + 3, -1, dummy_store);
261
262/*
263 * IN:
264 *  v0..3: byte-sliced 32-bit integers
265 * OUT:
266 *  v0..3: (IN <<< 1)
267 */
268#define rol32_1_16(v0, v1, v2, v3, t0, t1, t2, zero) \
269	vpcmpgtb v0, zero, t0; \
270	vpaddb v0, v0, v0; \
271	vpabsb t0, t0; \
272	\
273	vpcmpgtb v1, zero, t1; \
274	vpaddb v1, v1, v1; \
275	vpabsb t1, t1; \
276	\
277	vpcmpgtb v2, zero, t2; \
278	vpaddb v2, v2, v2; \
279	vpabsb t2, t2; \
280	\
281	vpor t0, v1, v1; \
282	\
283	vpcmpgtb v3, zero, t0; \
284	vpaddb v3, v3, v3; \
285	vpabsb t0, t0; \
286	\
287	vpor t1, v2, v2; \
288	vpor t2, v3, v3; \
289	vpor t0, v0, v0;
290
291/*
292 * IN:
293 *   r: byte-sliced AB state in memory
294 *   l: byte-sliced CD state in memory
295 * OUT:
296 *   x0..x7: new byte-sliced CD state
297 */
298#define fls16(l, l0, l1, l2, l3, l4, l5, l6, l7, r, t0, t1, t2, t3, tt0, \
299	      tt1, tt2, tt3, kll, klr, krl, krr) \
300	/* \
301	 * t0 = kll; \
302	 * t0 &= ll; \
303	 * lr ^= rol32(t0, 1); \
304	 */ \
305	vpxor tt0, tt0, tt0; \
306	vmovd kll, t0; \
307	vpshufb tt0, t0, t3; \
308	vpsrldq $1, t0, t0; \
309	vpshufb tt0, t0, t2; \
310	vpsrldq $1, t0, t0; \
311	vpshufb tt0, t0, t1; \
312	vpsrldq $1, t0, t0; \
313	vpshufb tt0, t0, t0; \
314	\
315	vpand l0, t0, t0; \
316	vpand l1, t1, t1; \
317	vpand l2, t2, t2; \
318	vpand l3, t3, t3; \
319	\
320	rol32_1_16(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
321	\
322	vpxor l4, t0, l4; \
323	vmovdqu l4, 4 * 16(l); \
324	vpxor l5, t1, l5; \
325	vmovdqu l5, 5 * 16(l); \
326	vpxor l6, t2, l6; \
327	vmovdqu l6, 6 * 16(l); \
328	vpxor l7, t3, l7; \
329	vmovdqu l7, 7 * 16(l); \
330	\
331	/* \
332	 * t2 = krr; \
333	 * t2 |= rr; \
334	 * rl ^= t2; \
335	 */ \
336	\
337	vmovd krr, t0; \
338	vpshufb tt0, t0, t3; \
339	vpsrldq $1, t0, t0; \
340	vpshufb tt0, t0, t2; \
341	vpsrldq $1, t0, t0; \
342	vpshufb tt0, t0, t1; \
343	vpsrldq $1, t0, t0; \
344	vpshufb tt0, t0, t0; \
345	\
346	vpor 4 * 16(r), t0, t0; \
347	vpor 5 * 16(r), t1, t1; \
348	vpor 6 * 16(r), t2, t2; \
349	vpor 7 * 16(r), t3, t3; \
350	\
351	vpxor 0 * 16(r), t0, t0; \
352	vpxor 1 * 16(r), t1, t1; \
353	vpxor 2 * 16(r), t2, t2; \
354	vpxor 3 * 16(r), t3, t3; \
355	vmovdqu t0, 0 * 16(r); \
356	vmovdqu t1, 1 * 16(r); \
357	vmovdqu t2, 2 * 16(r); \
358	vmovdqu t3, 3 * 16(r); \
359	\
360	/* \
361	 * t2 = krl; \
362	 * t2 &= rl; \
363	 * rr ^= rol32(t2, 1); \
364	 */ \
365	vmovd krl, t0; \
366	vpshufb tt0, t0, t3; \
367	vpsrldq $1, t0, t0; \
368	vpshufb tt0, t0, t2; \
369	vpsrldq $1, t0, t0; \
370	vpshufb tt0, t0, t1; \
371	vpsrldq $1, t0, t0; \
372	vpshufb tt0, t0, t0; \
373	\
374	vpand 0 * 16(r), t0, t0; \
375	vpand 1 * 16(r), t1, t1; \
376	vpand 2 * 16(r), t2, t2; \
377	vpand 3 * 16(r), t3, t3; \
378	\
379	rol32_1_16(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
380	\
381	vpxor 4 * 16(r), t0, t0; \
382	vpxor 5 * 16(r), t1, t1; \
383	vpxor 6 * 16(r), t2, t2; \
384	vpxor 7 * 16(r), t3, t3; \
385	vmovdqu t0, 4 * 16(r); \
386	vmovdqu t1, 5 * 16(r); \
387	vmovdqu t2, 6 * 16(r); \
388	vmovdqu t3, 7 * 16(r); \
389	\
390	/* \
391	 * t0 = klr; \
392	 * t0 |= lr; \
393	 * ll ^= t0; \
394	 */ \
395	\
396	vmovd klr, t0; \
397	vpshufb tt0, t0, t3; \
398	vpsrldq $1, t0, t0; \
399	vpshufb tt0, t0, t2; \
400	vpsrldq $1, t0, t0; \
401	vpshufb tt0, t0, t1; \
402	vpsrldq $1, t0, t0; \
403	vpshufb tt0, t0, t0; \
404	\
405	vpor l4, t0, t0; \
406	vpor l5, t1, t1; \
407	vpor l6, t2, t2; \
408	vpor l7, t3, t3; \
409	\
410	vpxor l0, t0, l0; \
411	vmovdqu l0, 0 * 16(l); \
412	vpxor l1, t1, l1; \
413	vmovdqu l1, 1 * 16(l); \
414	vpxor l2, t2, l2; \
415	vmovdqu l2, 2 * 16(l); \
416	vpxor l3, t3, l3; \
417	vmovdqu l3, 3 * 16(l);
418
419#define transpose_4x4(x0, x1, x2, x3, t1, t2) \
420	vpunpckhdq x1, x0, t2; \
421	vpunpckldq x1, x0, x0; \
422	\
423	vpunpckldq x3, x2, t1; \
424	vpunpckhdq x3, x2, x2; \
425	\
426	vpunpckhqdq t1, x0, x1; \
427	vpunpcklqdq t1, x0, x0; \
428	\
429	vpunpckhqdq x2, t2, x3; \
430	vpunpcklqdq x2, t2, x2;
431
432#define byteslice_16x16b(a0, b0, c0, d0, a1, b1, c1, d1, a2, b2, c2, d2, a3, \
433			 b3, c3, d3, st0, st1) \
434	vmovdqu d2, st0; \
435	vmovdqu d3, st1; \
436	transpose_4x4(a0, a1, a2, a3, d2, d3); \
437	transpose_4x4(b0, b1, b2, b3, d2, d3); \
438	vmovdqu st0, d2; \
439	vmovdqu st1, d3; \
440	\
441	vmovdqu a0, st0; \
442	vmovdqu a1, st1; \
443	transpose_4x4(c0, c1, c2, c3, a0, a1); \
444	transpose_4x4(d0, d1, d2, d3, a0, a1); \
445	\
446	vmovdqu .Lshufb_16x16b(%rip), a0; \
447	vmovdqu st1, a1; \
448	vpshufb a0, a2, a2; \
449	vpshufb a0, a3, a3; \
450	vpshufb a0, b0, b0; \
451	vpshufb a0, b1, b1; \
452	vpshufb a0, b2, b2; \
453	vpshufb a0, b3, b3; \
454	vpshufb a0, a1, a1; \
455	vpshufb a0, c0, c0; \
456	vpshufb a0, c1, c1; \
457	vpshufb a0, c2, c2; \
458	vpshufb a0, c3, c3; \
459	vpshufb a0, d0, d0; \
460	vpshufb a0, d1, d1; \
461	vpshufb a0, d2, d2; \
462	vpshufb a0, d3, d3; \
463	vmovdqu d3, st1; \
464	vmovdqu st0, d3; \
465	vpshufb a0, d3, a0; \
466	vmovdqu d2, st0; \
467	\
468	transpose_4x4(a0, b0, c0, d0, d2, d3); \
469	transpose_4x4(a1, b1, c1, d1, d2, d3); \
470	vmovdqu st0, d2; \
471	vmovdqu st1, d3; \
472	\
473	vmovdqu b0, st0; \
474	vmovdqu b1, st1; \
475	transpose_4x4(a2, b2, c2, d2, b0, b1); \
476	transpose_4x4(a3, b3, c3, d3, b0, b1); \
477	vmovdqu st0, b0; \
478	vmovdqu st1, b1; \
479	/* does not adjust output bytes inside vectors */
480
481/* load blocks to registers and apply pre-whitening */
482#define inpack16_pre(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
483		     y6, y7, rio, key) \
484	vmovq key, x0; \
485	vpshufb .Lpack_bswap(%rip), x0, x0; \
486	\
487	vpxor 0 * 16(rio), x0, y7; \
488	vpxor 1 * 16(rio), x0, y6; \
489	vpxor 2 * 16(rio), x0, y5; \
490	vpxor 3 * 16(rio), x0, y4; \
491	vpxor 4 * 16(rio), x0, y3; \
492	vpxor 5 * 16(rio), x0, y2; \
493	vpxor 6 * 16(rio), x0, y1; \
494	vpxor 7 * 16(rio), x0, y0; \
495	vpxor 8 * 16(rio), x0, x7; \
496	vpxor 9 * 16(rio), x0, x6; \
497	vpxor 10 * 16(rio), x0, x5; \
498	vpxor 11 * 16(rio), x0, x4; \
499	vpxor 12 * 16(rio), x0, x3; \
500	vpxor 13 * 16(rio), x0, x2; \
501	vpxor 14 * 16(rio), x0, x1; \
502	vpxor 15 * 16(rio), x0, x0;
503
504/* byteslice pre-whitened blocks and store to temporary memory */
505#define inpack16_post(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
506		      y6, y7, mem_ab, mem_cd) \
507	byteslice_16x16b(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, \
508			 y5, y6, y7, (mem_ab), (mem_cd)); \
509	\
510	vmovdqu x0, 0 * 16(mem_ab); \
511	vmovdqu x1, 1 * 16(mem_ab); \
512	vmovdqu x2, 2 * 16(mem_ab); \
513	vmovdqu x3, 3 * 16(mem_ab); \
514	vmovdqu x4, 4 * 16(mem_ab); \
515	vmovdqu x5, 5 * 16(mem_ab); \
516	vmovdqu x6, 6 * 16(mem_ab); \
517	vmovdqu x7, 7 * 16(mem_ab); \
518	vmovdqu y0, 0 * 16(mem_cd); \
519	vmovdqu y1, 1 * 16(mem_cd); \
520	vmovdqu y2, 2 * 16(mem_cd); \
521	vmovdqu y3, 3 * 16(mem_cd); \
522	vmovdqu y4, 4 * 16(mem_cd); \
523	vmovdqu y5, 5 * 16(mem_cd); \
524	vmovdqu y6, 6 * 16(mem_cd); \
525	vmovdqu y7, 7 * 16(mem_cd);
526
527/* de-byteslice, apply post-whitening and store blocks */
528#define outunpack16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, \
529		    y5, y6, y7, key, stack_tmp0, stack_tmp1) \
530	byteslice_16x16b(y0, y4, x0, x4, y1, y5, x1, x5, y2, y6, x2, x6, y3, \
531			 y7, x3, x7, stack_tmp0, stack_tmp1); \
532	\
533	vmovdqu x0, stack_tmp0; \
534	\
535	vmovq key, x0; \
536	vpshufb .Lpack_bswap(%rip), x0, x0; \
537	\
538	vpxor x0, y7, y7; \
539	vpxor x0, y6, y6; \
540	vpxor x0, y5, y5; \
541	vpxor x0, y4, y4; \
542	vpxor x0, y3, y3; \
543	vpxor x0, y2, y2; \
544	vpxor x0, y1, y1; \
545	vpxor x0, y0, y0; \
546	vpxor x0, x7, x7; \
547	vpxor x0, x6, x6; \
548	vpxor x0, x5, x5; \
549	vpxor x0, x4, x4; \
550	vpxor x0, x3, x3; \
551	vpxor x0, x2, x2; \
552	vpxor x0, x1, x1; \
553	vpxor stack_tmp0, x0, x0;
554
555#define write_output(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
556		     y6, y7, rio) \
557	vmovdqu x0, 0 * 16(rio); \
558	vmovdqu x1, 1 * 16(rio); \
559	vmovdqu x2, 2 * 16(rio); \
560	vmovdqu x3, 3 * 16(rio); \
561	vmovdqu x4, 4 * 16(rio); \
562	vmovdqu x5, 5 * 16(rio); \
563	vmovdqu x6, 6 * 16(rio); \
564	vmovdqu x7, 7 * 16(rio); \
565	vmovdqu y0, 8 * 16(rio); \
566	vmovdqu y1, 9 * 16(rio); \
567	vmovdqu y2, 10 * 16(rio); \
568	vmovdqu y3, 11 * 16(rio); \
569	vmovdqu y4, 12 * 16(rio); \
570	vmovdqu y5, 13 * 16(rio); \
571	vmovdqu y6, 14 * 16(rio); \
572	vmovdqu y7, 15 * 16(rio);
573
574
575/* NB: section is mergeable, all elements must be aligned 16-byte blocks */
576.section	.rodata.cst16, "aM", @progbits, 16
577.align 16
578
579#define SHUFB_BYTES(idx) \
580	0 + (idx), 4 + (idx), 8 + (idx), 12 + (idx)
581
582.Lshufb_16x16b:
583	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3);
584
585.Lpack_bswap:
586	.long 0x00010203
587	.long 0x04050607
588	.long 0x80808080
589	.long 0x80808080
590
591/*
592 * pre-SubByte transform
593 *
594 * pre-lookup for sbox1, sbox2, sbox3:
595 *   swap_bitendianness(
596 *       isom_map_camellia_to_aes(
597 *           camellia_f(
598 *               swap_bitendianess(in)
599 *           )
600 *       )
601 *   )
602 *
603 * (note: '⊕ 0xc5' inside camellia_f())
604 */
605.Lpre_tf_lo_s1:
606	.byte 0x45, 0xe8, 0x40, 0xed, 0x2e, 0x83, 0x2b, 0x86
607	.byte 0x4b, 0xe6, 0x4e, 0xe3, 0x20, 0x8d, 0x25, 0x88
608.Lpre_tf_hi_s1:
609	.byte 0x00, 0x51, 0xf1, 0xa0, 0x8a, 0xdb, 0x7b, 0x2a
610	.byte 0x09, 0x58, 0xf8, 0xa9, 0x83, 0xd2, 0x72, 0x23
611
612/*
613 * pre-SubByte transform
614 *
615 * pre-lookup for sbox4:
616 *   swap_bitendianness(
617 *       isom_map_camellia_to_aes(
618 *           camellia_f(
619 *               swap_bitendianess(in <<< 1)
620 *           )
621 *       )
622 *   )
623 *
624 * (note: '⊕ 0xc5' inside camellia_f())
625 */
626.Lpre_tf_lo_s4:
627	.byte 0x45, 0x40, 0x2e, 0x2b, 0x4b, 0x4e, 0x20, 0x25
628	.byte 0x14, 0x11, 0x7f, 0x7a, 0x1a, 0x1f, 0x71, 0x74
629.Lpre_tf_hi_s4:
630	.byte 0x00, 0xf1, 0x8a, 0x7b, 0x09, 0xf8, 0x83, 0x72
631	.byte 0xad, 0x5c, 0x27, 0xd6, 0xa4, 0x55, 0x2e, 0xdf
632
633/*
634 * post-SubByte transform
635 *
636 * post-lookup for sbox1, sbox4:
637 *  swap_bitendianness(
638 *      camellia_h(
639 *          isom_map_aes_to_camellia(
640 *              swap_bitendianness(
641 *                  aes_inverse_affine_transform(in)
642 *              )
643 *          )
644 *      )
645 *  )
646 *
647 * (note: '⊕ 0x6e' inside camellia_h())
648 */
649.Lpost_tf_lo_s1:
650	.byte 0x3c, 0xcc, 0xcf, 0x3f, 0x32, 0xc2, 0xc1, 0x31
651	.byte 0xdc, 0x2c, 0x2f, 0xdf, 0xd2, 0x22, 0x21, 0xd1
652.Lpost_tf_hi_s1:
653	.byte 0x00, 0xf9, 0x86, 0x7f, 0xd7, 0x2e, 0x51, 0xa8
654	.byte 0xa4, 0x5d, 0x22, 0xdb, 0x73, 0x8a, 0xf5, 0x0c
655
656/*
657 * post-SubByte transform
658 *
659 * post-lookup for sbox2:
660 *  swap_bitendianness(
661 *      camellia_h(
662 *          isom_map_aes_to_camellia(
663 *              swap_bitendianness(
664 *                  aes_inverse_affine_transform(in)
665 *              )
666 *          )
667 *      )
668 *  ) <<< 1
669 *
670 * (note: '⊕ 0x6e' inside camellia_h())
671 */
672.Lpost_tf_lo_s2:
673	.byte 0x78, 0x99, 0x9f, 0x7e, 0x64, 0x85, 0x83, 0x62
674	.byte 0xb9, 0x58, 0x5e, 0xbf, 0xa5, 0x44, 0x42, 0xa3
675.Lpost_tf_hi_s2:
676	.byte 0x00, 0xf3, 0x0d, 0xfe, 0xaf, 0x5c, 0xa2, 0x51
677	.byte 0x49, 0xba, 0x44, 0xb7, 0xe6, 0x15, 0xeb, 0x18
678
679/*
680 * post-SubByte transform
681 *
682 * post-lookup for sbox3:
683 *  swap_bitendianness(
684 *      camellia_h(
685 *          isom_map_aes_to_camellia(
686 *              swap_bitendianness(
687 *                  aes_inverse_affine_transform(in)
688 *              )
689 *          )
690 *      )
691 *  ) >>> 1
692 *
693 * (note: '⊕ 0x6e' inside camellia_h())
694 */
695.Lpost_tf_lo_s3:
696	.byte 0x1e, 0x66, 0xe7, 0x9f, 0x19, 0x61, 0xe0, 0x98
697	.byte 0x6e, 0x16, 0x97, 0xef, 0x69, 0x11, 0x90, 0xe8
698.Lpost_tf_hi_s3:
699	.byte 0x00, 0xfc, 0x43, 0xbf, 0xeb, 0x17, 0xa8, 0x54
700	.byte 0x52, 0xae, 0x11, 0xed, 0xb9, 0x45, 0xfa, 0x06
701
702/* For isolating SubBytes from AESENCLAST, inverse shift row */
703.Linv_shift_row:
704	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
705	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
706
707/* 4-bit mask */
708.section	.rodata.cst4.L0f0f0f0f, "aM", @progbits, 4
709.align 4
710.L0f0f0f0f:
711	.long 0x0f0f0f0f
712
713.text
714
715SYM_FUNC_START_LOCAL(__camellia_enc_blk16)
716	/* input:
717	 *	%rdi: ctx, CTX
718	 *	%rax: temporary storage, 256 bytes
719	 *	%xmm0..%xmm15: 16 plaintext blocks
720	 * output:
721	 *	%xmm0..%xmm15: 16 encrypted blocks, order swapped:
722	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
723	 */
724	FRAME_BEGIN
725
726	leaq 8 * 16(%rax), %rcx;
727
728	inpack16_post(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
729		      %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
730		      %xmm15, %rax, %rcx);
731
732	enc_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
733		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
734		     %xmm15, %rax, %rcx, 0);
735
736	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
737	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
738	      %xmm15,
739	      ((key_table + (8) * 8) + 0)(CTX),
740	      ((key_table + (8) * 8) + 4)(CTX),
741	      ((key_table + (8) * 8) + 8)(CTX),
742	      ((key_table + (8) * 8) + 12)(CTX));
743
744	enc_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
745		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
746		     %xmm15, %rax, %rcx, 8);
747
748	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
749	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
750	      %xmm15,
751	      ((key_table + (16) * 8) + 0)(CTX),
752	      ((key_table + (16) * 8) + 4)(CTX),
753	      ((key_table + (16) * 8) + 8)(CTX),
754	      ((key_table + (16) * 8) + 12)(CTX));
755
756	enc_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
757		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
758		     %xmm15, %rax, %rcx, 16);
759
760	movl $24, %r8d;
761	cmpl $16, key_length(CTX);
762	jne .Lenc_max32;
763
764.Lenc_done:
765	/* load CD for output */
766	vmovdqu 0 * 16(%rcx), %xmm8;
767	vmovdqu 1 * 16(%rcx), %xmm9;
768	vmovdqu 2 * 16(%rcx), %xmm10;
769	vmovdqu 3 * 16(%rcx), %xmm11;
770	vmovdqu 4 * 16(%rcx), %xmm12;
771	vmovdqu 5 * 16(%rcx), %xmm13;
772	vmovdqu 6 * 16(%rcx), %xmm14;
773	vmovdqu 7 * 16(%rcx), %xmm15;
774
775	outunpack16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
776		    %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
777		    %xmm15, (key_table)(CTX, %r8, 8), (%rax), 1 * 16(%rax));
778
779	FRAME_END
780	RET;
781
782.align 8
783.Lenc_max32:
784	movl $32, %r8d;
785
786	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
787	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
788	      %xmm15,
789	      ((key_table + (24) * 8) + 0)(CTX),
790	      ((key_table + (24) * 8) + 4)(CTX),
791	      ((key_table + (24) * 8) + 8)(CTX),
792	      ((key_table + (24) * 8) + 12)(CTX));
793
794	enc_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
795		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
796		     %xmm15, %rax, %rcx, 24);
797
798	jmp .Lenc_done;
799SYM_FUNC_END(__camellia_enc_blk16)
800
801SYM_FUNC_START_LOCAL(__camellia_dec_blk16)
802	/* input:
803	 *	%rdi: ctx, CTX
804	 *	%rax: temporary storage, 256 bytes
805	 *	%r8d: 24 for 16 byte key, 32 for larger
806	 *	%xmm0..%xmm15: 16 encrypted blocks
807	 * output:
808	 *	%xmm0..%xmm15: 16 plaintext blocks, order swapped:
809	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
810	 */
811	FRAME_BEGIN
812
813	leaq 8 * 16(%rax), %rcx;
814
815	inpack16_post(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
816		      %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
817		      %xmm15, %rax, %rcx);
818
819	cmpl $32, %r8d;
820	je .Ldec_max32;
821
822.Ldec_max24:
823	dec_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
824		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
825		     %xmm15, %rax, %rcx, 16);
826
827	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
828	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
829	      %xmm15,
830	      ((key_table + (16) * 8) + 8)(CTX),
831	      ((key_table + (16) * 8) + 12)(CTX),
832	      ((key_table + (16) * 8) + 0)(CTX),
833	      ((key_table + (16) * 8) + 4)(CTX));
834
835	dec_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
836		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
837		     %xmm15, %rax, %rcx, 8);
838
839	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
840	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
841	      %xmm15,
842	      ((key_table + (8) * 8) + 8)(CTX),
843	      ((key_table + (8) * 8) + 12)(CTX),
844	      ((key_table + (8) * 8) + 0)(CTX),
845	      ((key_table + (8) * 8) + 4)(CTX));
846
847	dec_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
848		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
849		     %xmm15, %rax, %rcx, 0);
850
851	/* load CD for output */
852	vmovdqu 0 * 16(%rcx), %xmm8;
853	vmovdqu 1 * 16(%rcx), %xmm9;
854	vmovdqu 2 * 16(%rcx), %xmm10;
855	vmovdqu 3 * 16(%rcx), %xmm11;
856	vmovdqu 4 * 16(%rcx), %xmm12;
857	vmovdqu 5 * 16(%rcx), %xmm13;
858	vmovdqu 6 * 16(%rcx), %xmm14;
859	vmovdqu 7 * 16(%rcx), %xmm15;
860
861	outunpack16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
862		    %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
863		    %xmm15, (key_table)(CTX), (%rax), 1 * 16(%rax));
864
865	FRAME_END
866	RET;
867
868.align 8
869.Ldec_max32:
870	dec_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
871		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
872		     %xmm15, %rax, %rcx, 24);
873
874	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
875	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
876	      %xmm15,
877	      ((key_table + (24) * 8) + 8)(CTX),
878	      ((key_table + (24) * 8) + 12)(CTX),
879	      ((key_table + (24) * 8) + 0)(CTX),
880	      ((key_table + (24) * 8) + 4)(CTX));
881
882	jmp .Ldec_max24;
883SYM_FUNC_END(__camellia_dec_blk16)
884
885SYM_FUNC_START(camellia_ecb_enc_16way)
886	/* input:
887	 *	%rdi: ctx, CTX
888	 *	%rsi: dst (16 blocks)
889	 *	%rdx: src (16 blocks)
890	 */
891	 FRAME_BEGIN
892
893	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
894		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
895		     %xmm15, %rdx, (key_table)(CTX));
896
897	/* now dst can be used as temporary buffer (even in src == dst case) */
898	movq	%rsi, %rax;
899
900	call __camellia_enc_blk16;
901
902	write_output(%xmm7, %xmm6, %xmm5, %xmm4, %xmm3, %xmm2, %xmm1, %xmm0,
903		     %xmm15, %xmm14, %xmm13, %xmm12, %xmm11, %xmm10, %xmm9,
904		     %xmm8, %rsi);
905
906	FRAME_END
907	RET;
908SYM_FUNC_END(camellia_ecb_enc_16way)
909
910SYM_FUNC_START(camellia_ecb_dec_16way)
911	/* input:
912	 *	%rdi: ctx, CTX
913	 *	%rsi: dst (16 blocks)
914	 *	%rdx: src (16 blocks)
915	 */
916	 FRAME_BEGIN
917
918	cmpl $16, key_length(CTX);
919	movl $32, %r8d;
920	movl $24, %eax;
921	cmovel %eax, %r8d; /* max */
922
923	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
924		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
925		     %xmm15, %rdx, (key_table)(CTX, %r8, 8));
926
927	/* now dst can be used as temporary buffer (even in src == dst case) */
928	movq	%rsi, %rax;
929
930	call __camellia_dec_blk16;
931
932	write_output(%xmm7, %xmm6, %xmm5, %xmm4, %xmm3, %xmm2, %xmm1, %xmm0,
933		     %xmm15, %xmm14, %xmm13, %xmm12, %xmm11, %xmm10, %xmm9,
934		     %xmm8, %rsi);
935
936	FRAME_END
937	RET;
938SYM_FUNC_END(camellia_ecb_dec_16way)
939
940SYM_FUNC_START(camellia_cbc_dec_16way)
941	/* input:
942	 *	%rdi: ctx, CTX
943	 *	%rsi: dst (16 blocks)
944	 *	%rdx: src (16 blocks)
945	 */
946	FRAME_BEGIN
947
948	cmpl $16, key_length(CTX);
949	movl $32, %r8d;
950	movl $24, %eax;
951	cmovel %eax, %r8d; /* max */
952
953	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
954		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
955		     %xmm15, %rdx, (key_table)(CTX, %r8, 8));
956
957	/*
958	 * dst might still be in-use (in case dst == src), so use stack for
959	 * temporary storage.
960	 */
961	subq $(16 * 16), %rsp;
962	movq %rsp, %rax;
963
964	call __camellia_dec_blk16;
965
966	addq $(16 * 16), %rsp;
967
968	vpxor (0 * 16)(%rdx), %xmm6, %xmm6;
969	vpxor (1 * 16)(%rdx), %xmm5, %xmm5;
970	vpxor (2 * 16)(%rdx), %xmm4, %xmm4;
971	vpxor (3 * 16)(%rdx), %xmm3, %xmm3;
972	vpxor (4 * 16)(%rdx), %xmm2, %xmm2;
973	vpxor (5 * 16)(%rdx), %xmm1, %xmm1;
974	vpxor (6 * 16)(%rdx), %xmm0, %xmm0;
975	vpxor (7 * 16)(%rdx), %xmm15, %xmm15;
976	vpxor (8 * 16)(%rdx), %xmm14, %xmm14;
977	vpxor (9 * 16)(%rdx), %xmm13, %xmm13;
978	vpxor (10 * 16)(%rdx), %xmm12, %xmm12;
979	vpxor (11 * 16)(%rdx), %xmm11, %xmm11;
980	vpxor (12 * 16)(%rdx), %xmm10, %xmm10;
981	vpxor (13 * 16)(%rdx), %xmm9, %xmm9;
982	vpxor (14 * 16)(%rdx), %xmm8, %xmm8;
983	write_output(%xmm7, %xmm6, %xmm5, %xmm4, %xmm3, %xmm2, %xmm1, %xmm0,
984		     %xmm15, %xmm14, %xmm13, %xmm12, %xmm11, %xmm10, %xmm9,
985		     %xmm8, %rsi);
986
987	FRAME_END
988	RET;
989SYM_FUNC_END(camellia_cbc_dec_16way)
990