1/*
2 * x86_64/AVX2/AES-NI assembler implementation of Camellia
3 *
4 * Copyright © 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#include <linux/linkage.h>
14#include <asm/frame.h>
15#include <asm/nospec-branch.h>
16
17#define CAMELLIA_TABLE_BYTE_LEN 272
18
19/* struct camellia_ctx: */
20#define key_table 0
21#define key_length CAMELLIA_TABLE_BYTE_LEN
22
23/* register macros */
24#define CTX %rdi
25#define RIO %r8
26
27/**********************************************************************
28  helper macros
29 **********************************************************************/
30#define filter_8bit(x, lo_t, hi_t, mask4bit, tmp0) \
31	vpand x, mask4bit, tmp0; \
32	vpandn x, mask4bit, x; \
33	vpsrld $4, x, x; \
34	\
35	vpshufb tmp0, lo_t, tmp0; \
36	vpshufb x, hi_t, x; \
37	vpxor tmp0, x, x;
38
39#define ymm0_x xmm0
40#define ymm1_x xmm1
41#define ymm2_x xmm2
42#define ymm3_x xmm3
43#define ymm4_x xmm4
44#define ymm5_x xmm5
45#define ymm6_x xmm6
46#define ymm7_x xmm7
47#define ymm8_x xmm8
48#define ymm9_x xmm9
49#define ymm10_x xmm10
50#define ymm11_x xmm11
51#define ymm12_x xmm12
52#define ymm13_x xmm13
53#define ymm14_x xmm14
54#define ymm15_x xmm15
55
56/**********************************************************************
57  32-way camellia
58 **********************************************************************/
59
60/*
61 * IN:
62 *   x0..x7: byte-sliced AB state
63 *   mem_cd: register pointer storing CD state
64 *   key: index for key material
65 * OUT:
66 *   x0..x7: new byte-sliced CD state
67 */
68#define roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, t0, t1, t2, t3, t4, t5, t6, \
69		  t7, mem_cd, key) \
70	/* \
71	 * S-function with AES subbytes \
72	 */ \
73	vbroadcasti128 .Linv_shift_row, t4; \
74	vpbroadcastd .L0f0f0f0f, t7; \
75	vbroadcasti128 .Lpre_tf_lo_s1, t5; \
76	vbroadcasti128 .Lpre_tf_hi_s1, t6; \
77	vbroadcasti128 .Lpre_tf_lo_s4, t2; \
78	vbroadcasti128 .Lpre_tf_hi_s4, t3; \
79	\
80	/* AES inverse shift rows */ \
81	vpshufb t4, x0, x0; \
82	vpshufb t4, x7, x7; \
83	vpshufb t4, x3, x3; \
84	vpshufb t4, x6, x6; \
85	vpshufb t4, x2, x2; \
86	vpshufb t4, x5, x5; \
87	vpshufb t4, x1, x1; \
88	vpshufb t4, x4, x4; \
89	\
90	/* prefilter sboxes 1, 2 and 3 */ \
91	/* prefilter sbox 4 */ \
92	filter_8bit(x0, t5, t6, t7, t4); \
93	filter_8bit(x7, t5, t6, t7, t4); \
94	vextracti128 $1, x0, t0##_x; \
95	vextracti128 $1, x7, t1##_x; \
96	filter_8bit(x3, t2, t3, t7, t4); \
97	filter_8bit(x6, t2, t3, t7, t4); \
98	vextracti128 $1, x3, t3##_x; \
99	vextracti128 $1, x6, t2##_x; \
100	filter_8bit(x2, t5, t6, t7, t4); \
101	filter_8bit(x5, t5, t6, t7, t4); \
102	filter_8bit(x1, t5, t6, t7, t4); \
103	filter_8bit(x4, t5, t6, t7, t4); \
104	\
105	vpxor t4##_x, t4##_x, t4##_x; \
106	\
107	/* AES subbytes + AES shift rows */ \
108	vextracti128 $1, x2, t6##_x; \
109	vextracti128 $1, x5, t5##_x; \
110	vaesenclast t4##_x, x0##_x, x0##_x; \
111	vaesenclast t4##_x, t0##_x, t0##_x; \
112	vinserti128 $1, t0##_x, x0, x0; \
113	vaesenclast t4##_x, x7##_x, x7##_x; \
114	vaesenclast t4##_x, t1##_x, t1##_x; \
115	vinserti128 $1, t1##_x, x7, x7; \
116	vaesenclast t4##_x, x3##_x, x3##_x; \
117	vaesenclast t4##_x, t3##_x, t3##_x; \
118	vinserti128 $1, t3##_x, x3, x3; \
119	vaesenclast t4##_x, x6##_x, x6##_x; \
120	vaesenclast t4##_x, t2##_x, t2##_x; \
121	vinserti128 $1, t2##_x, x6, x6; \
122	vextracti128 $1, x1, t3##_x; \
123	vextracti128 $1, x4, t2##_x; \
124	vbroadcasti128 .Lpost_tf_lo_s1, t0; \
125	vbroadcasti128 .Lpost_tf_hi_s1, t1; \
126	vaesenclast t4##_x, x2##_x, x2##_x; \
127	vaesenclast t4##_x, t6##_x, t6##_x; \
128	vinserti128 $1, t6##_x, x2, x2; \
129	vaesenclast t4##_x, x5##_x, x5##_x; \
130	vaesenclast t4##_x, t5##_x, t5##_x; \
131	vinserti128 $1, t5##_x, x5, x5; \
132	vaesenclast t4##_x, x1##_x, x1##_x; \
133	vaesenclast t4##_x, t3##_x, t3##_x; \
134	vinserti128 $1, t3##_x, x1, x1; \
135	vaesenclast t4##_x, x4##_x, x4##_x; \
136	vaesenclast t4##_x, t2##_x, t2##_x; \
137	vinserti128 $1, t2##_x, x4, x4; \
138	\
139	/* postfilter sboxes 1 and 4 */ \
140	vbroadcasti128 .Lpost_tf_lo_s3, t2; \
141	vbroadcasti128 .Lpost_tf_hi_s3, t3; \
142	filter_8bit(x0, t0, t1, t7, t6); \
143	filter_8bit(x7, t0, t1, t7, t6); \
144	filter_8bit(x3, t0, t1, t7, t6); \
145	filter_8bit(x6, t0, t1, t7, t6); \
146	\
147	/* postfilter sbox 3 */ \
148	vbroadcasti128 .Lpost_tf_lo_s2, t4; \
149	vbroadcasti128 .Lpost_tf_hi_s2, t5; \
150	filter_8bit(x2, t2, t3, t7, t6); \
151	filter_8bit(x5, t2, t3, t7, t6); \
152	\
153	vpbroadcastq key, t0; /* higher 64-bit duplicate ignored */ \
154	\
155	/* postfilter sbox 2 */ \
156	filter_8bit(x1, t4, t5, t7, t2); \
157	filter_8bit(x4, t4, t5, t7, t2); \
158	vpxor t7, t7, t7; \
159	\
160	vpsrldq $1, t0, t1; \
161	vpsrldq $2, t0, t2; \
162	vpshufb t7, t1, t1; \
163	vpsrldq $3, t0, t3; \
164	\
165	/* P-function */ \
166	vpxor x5, x0, x0; \
167	vpxor x6, x1, x1; \
168	vpxor x7, x2, x2; \
169	vpxor x4, x3, x3; \
170	\
171	vpshufb t7, t2, t2; \
172	vpsrldq $4, t0, t4; \
173	vpshufb t7, t3, t3; \
174	vpsrldq $5, t0, t5; \
175	vpshufb t7, t4, t4; \
176	\
177	vpxor x2, x4, x4; \
178	vpxor x3, x5, x5; \
179	vpxor x0, x6, x6; \
180	vpxor x1, x7, x7; \
181	\
182	vpsrldq $6, t0, t6; \
183	vpshufb t7, t5, t5; \
184	vpshufb t7, t6, t6; \
185	\
186	vpxor x7, x0, x0; \
187	vpxor x4, x1, x1; \
188	vpxor x5, x2, x2; \
189	vpxor x6, x3, x3; \
190	\
191	vpxor x3, x4, x4; \
192	vpxor x0, x5, x5; \
193	vpxor x1, x6, x6; \
194	vpxor x2, x7, x7; /* note: high and low parts swapped */ \
195	\
196	/* Add key material and result to CD (x becomes new CD) */ \
197	\
198	vpxor t6, x1, x1; \
199	vpxor 5 * 32(mem_cd), x1, x1; \
200	\
201	vpsrldq $7, t0, t6; \
202	vpshufb t7, t0, t0; \
203	vpshufb t7, t6, t7; \
204	\
205	vpxor t7, x0, x0; \
206	vpxor 4 * 32(mem_cd), x0, x0; \
207	\
208	vpxor t5, x2, x2; \
209	vpxor 6 * 32(mem_cd), x2, x2; \
210	\
211	vpxor t4, x3, x3; \
212	vpxor 7 * 32(mem_cd), x3, x3; \
213	\
214	vpxor t3, x4, x4; \
215	vpxor 0 * 32(mem_cd), x4, x4; \
216	\
217	vpxor t2, x5, x5; \
218	vpxor 1 * 32(mem_cd), x5, x5; \
219	\
220	vpxor t1, x6, x6; \
221	vpxor 2 * 32(mem_cd), x6, x6; \
222	\
223	vpxor t0, x7, x7; \
224	vpxor 3 * 32(mem_cd), x7, x7;
225
226/*
227 * Size optimization... with inlined roundsm32 binary would be over 5 times
228 * larger and would only marginally faster.
229 */
230.align 8
231roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd:
232	roundsm32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
233		  %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14, %ymm15,
234		  %rcx, (%r9));
235	ret;
236ENDPROC(roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
237
238.align 8
239roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab:
240	roundsm32(%ymm4, %ymm5, %ymm6, %ymm7, %ymm0, %ymm1, %ymm2, %ymm3,
241		  %ymm12, %ymm13, %ymm14, %ymm15, %ymm8, %ymm9, %ymm10, %ymm11,
242		  %rax, (%r9));
243	ret;
244ENDPROC(roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
245
246/*
247 * IN/OUT:
248 *  x0..x7: byte-sliced AB state preloaded
249 *  mem_ab: byte-sliced AB state in memory
250 *  mem_cb: byte-sliced CD state in memory
251 */
252#define two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
253		      y6, y7, mem_ab, mem_cd, i, dir, store_ab) \
254	leaq (key_table + (i) * 8)(CTX), %r9; \
255	call roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd; \
256	\
257	vmovdqu x0, 4 * 32(mem_cd); \
258	vmovdqu x1, 5 * 32(mem_cd); \
259	vmovdqu x2, 6 * 32(mem_cd); \
260	vmovdqu x3, 7 * 32(mem_cd); \
261	vmovdqu x4, 0 * 32(mem_cd); \
262	vmovdqu x5, 1 * 32(mem_cd); \
263	vmovdqu x6, 2 * 32(mem_cd); \
264	vmovdqu x7, 3 * 32(mem_cd); \
265	\
266	leaq (key_table + ((i) + (dir)) * 8)(CTX), %r9; \
267	call roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab; \
268	\
269	store_ab(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab);
270
271#define dummy_store(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) /* do nothing */
272
273#define store_ab_state(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) \
274	/* Store new AB state */ \
275	vmovdqu x4, 4 * 32(mem_ab); \
276	vmovdqu x5, 5 * 32(mem_ab); \
277	vmovdqu x6, 6 * 32(mem_ab); \
278	vmovdqu x7, 7 * 32(mem_ab); \
279	vmovdqu x0, 0 * 32(mem_ab); \
280	vmovdqu x1, 1 * 32(mem_ab); \
281	vmovdqu x2, 2 * 32(mem_ab); \
282	vmovdqu x3, 3 * 32(mem_ab);
283
284#define enc_rounds32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
285		      y6, y7, mem_ab, mem_cd, i) \
286	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
287		      y6, y7, mem_ab, mem_cd, (i) + 2, 1, store_ab_state); \
288	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
289		      y6, y7, mem_ab, mem_cd, (i) + 4, 1, store_ab_state); \
290	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
291		      y6, y7, mem_ab, mem_cd, (i) + 6, 1, dummy_store);
292
293#define dec_rounds32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
294		      y6, y7, mem_ab, mem_cd, i) \
295	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
296		      y6, y7, mem_ab, mem_cd, (i) + 7, -1, store_ab_state); \
297	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
298		      y6, y7, mem_ab, mem_cd, (i) + 5, -1, store_ab_state); \
299	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
300		      y6, y7, mem_ab, mem_cd, (i) + 3, -1, dummy_store);
301
302/*
303 * IN:
304 *  v0..3: byte-sliced 32-bit integers
305 * OUT:
306 *  v0..3: (IN <<< 1)
307 */
308#define rol32_1_32(v0, v1, v2, v3, t0, t1, t2, zero) \
309	vpcmpgtb v0, zero, t0; \
310	vpaddb v0, v0, v0; \
311	vpabsb t0, t0; \
312	\
313	vpcmpgtb v1, zero, t1; \
314	vpaddb v1, v1, v1; \
315	vpabsb t1, t1; \
316	\
317	vpcmpgtb v2, zero, t2; \
318	vpaddb v2, v2, v2; \
319	vpabsb t2, t2; \
320	\
321	vpor t0, v1, v1; \
322	\
323	vpcmpgtb v3, zero, t0; \
324	vpaddb v3, v3, v3; \
325	vpabsb t0, t0; \
326	\
327	vpor t1, v2, v2; \
328	vpor t2, v3, v3; \
329	vpor t0, v0, v0;
330
331/*
332 * IN:
333 *   r: byte-sliced AB state in memory
334 *   l: byte-sliced CD state in memory
335 * OUT:
336 *   x0..x7: new byte-sliced CD state
337 */
338#define fls32(l, l0, l1, l2, l3, l4, l5, l6, l7, r, t0, t1, t2, t3, tt0, \
339	      tt1, tt2, tt3, kll, klr, krl, krr) \
340	/* \
341	 * t0 = kll; \
342	 * t0 &= ll; \
343	 * lr ^= rol32(t0, 1); \
344	 */ \
345	vpbroadcastd kll, t0; /* only lowest 32-bit used */ \
346	vpxor tt0, tt0, tt0; \
347	vpshufb tt0, t0, t3; \
348	vpsrldq $1, t0, t0; \
349	vpshufb tt0, t0, t2; \
350	vpsrldq $1, t0, t0; \
351	vpshufb tt0, t0, t1; \
352	vpsrldq $1, t0, t0; \
353	vpshufb tt0, t0, t0; \
354	\
355	vpand l0, t0, t0; \
356	vpand l1, t1, t1; \
357	vpand l2, t2, t2; \
358	vpand l3, t3, t3; \
359	\
360	rol32_1_32(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
361	\
362	vpxor l4, t0, l4; \
363	vpbroadcastd krr, t0; /* only lowest 32-bit used */ \
364	vmovdqu l4, 4 * 32(l); \
365	vpxor l5, t1, l5; \
366	vmovdqu l5, 5 * 32(l); \
367	vpxor l6, t2, l6; \
368	vmovdqu l6, 6 * 32(l); \
369	vpxor l7, t3, l7; \
370	vmovdqu l7, 7 * 32(l); \
371	\
372	/* \
373	 * t2 = krr; \
374	 * t2 |= rr; \
375	 * rl ^= t2; \
376	 */ \
377	\
378	vpshufb tt0, t0, t3; \
379	vpsrldq $1, t0, t0; \
380	vpshufb tt0, t0, t2; \
381	vpsrldq $1, t0, t0; \
382	vpshufb tt0, t0, t1; \
383	vpsrldq $1, t0, t0; \
384	vpshufb tt0, t0, t0; \
385	\
386	vpor 4 * 32(r), t0, t0; \
387	vpor 5 * 32(r), t1, t1; \
388	vpor 6 * 32(r), t2, t2; \
389	vpor 7 * 32(r), t3, t3; \
390	\
391	vpxor 0 * 32(r), t0, t0; \
392	vpxor 1 * 32(r), t1, t1; \
393	vpxor 2 * 32(r), t2, t2; \
394	vpxor 3 * 32(r), t3, t3; \
395	vmovdqu t0, 0 * 32(r); \
396	vpbroadcastd krl, t0; /* only lowest 32-bit used */ \
397	vmovdqu t1, 1 * 32(r); \
398	vmovdqu t2, 2 * 32(r); \
399	vmovdqu t3, 3 * 32(r); \
400	\
401	/* \
402	 * t2 = krl; \
403	 * t2 &= rl; \
404	 * rr ^= rol32(t2, 1); \
405	 */ \
406	vpshufb tt0, t0, t3; \
407	vpsrldq $1, t0, t0; \
408	vpshufb tt0, t0, t2; \
409	vpsrldq $1, t0, t0; \
410	vpshufb tt0, t0, t1; \
411	vpsrldq $1, t0, t0; \
412	vpshufb tt0, t0, t0; \
413	\
414	vpand 0 * 32(r), t0, t0; \
415	vpand 1 * 32(r), t1, t1; \
416	vpand 2 * 32(r), t2, t2; \
417	vpand 3 * 32(r), t3, t3; \
418	\
419	rol32_1_32(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
420	\
421	vpxor 4 * 32(r), t0, t0; \
422	vpxor 5 * 32(r), t1, t1; \
423	vpxor 6 * 32(r), t2, t2; \
424	vpxor 7 * 32(r), t3, t3; \
425	vmovdqu t0, 4 * 32(r); \
426	vpbroadcastd klr, t0; /* only lowest 32-bit used */ \
427	vmovdqu t1, 5 * 32(r); \
428	vmovdqu t2, 6 * 32(r); \
429	vmovdqu t3, 7 * 32(r); \
430	\
431	/* \
432	 * t0 = klr; \
433	 * t0 |= lr; \
434	 * ll ^= t0; \
435	 */ \
436	\
437	vpshufb tt0, t0, t3; \
438	vpsrldq $1, t0, t0; \
439	vpshufb tt0, t0, t2; \
440	vpsrldq $1, t0, t0; \
441	vpshufb tt0, t0, t1; \
442	vpsrldq $1, t0, t0; \
443	vpshufb tt0, t0, t0; \
444	\
445	vpor l4, t0, t0; \
446	vpor l5, t1, t1; \
447	vpor l6, t2, t2; \
448	vpor l7, t3, t3; \
449	\
450	vpxor l0, t0, l0; \
451	vmovdqu l0, 0 * 32(l); \
452	vpxor l1, t1, l1; \
453	vmovdqu l1, 1 * 32(l); \
454	vpxor l2, t2, l2; \
455	vmovdqu l2, 2 * 32(l); \
456	vpxor l3, t3, l3; \
457	vmovdqu l3, 3 * 32(l);
458
459#define transpose_4x4(x0, x1, x2, x3, t1, t2) \
460	vpunpckhdq x1, x0, t2; \
461	vpunpckldq x1, x0, x0; \
462	\
463	vpunpckldq x3, x2, t1; \
464	vpunpckhdq x3, x2, x2; \
465	\
466	vpunpckhqdq t1, x0, x1; \
467	vpunpcklqdq t1, x0, x0; \
468	\
469	vpunpckhqdq x2, t2, x3; \
470	vpunpcklqdq x2, t2, x2;
471
472#define byteslice_16x16b_fast(a0, b0, c0, d0, a1, b1, c1, d1, a2, b2, c2, d2, \
473			      a3, b3, c3, d3, st0, st1) \
474	vmovdqu d2, st0; \
475	vmovdqu d3, st1; \
476	transpose_4x4(a0, a1, a2, a3, d2, d3); \
477	transpose_4x4(b0, b1, b2, b3, d2, d3); \
478	vmovdqu st0, d2; \
479	vmovdqu st1, d3; \
480	\
481	vmovdqu a0, st0; \
482	vmovdqu a1, st1; \
483	transpose_4x4(c0, c1, c2, c3, a0, a1); \
484	transpose_4x4(d0, d1, d2, d3, a0, a1); \
485	\
486	vbroadcasti128 .Lshufb_16x16b, a0; \
487	vmovdqu st1, a1; \
488	vpshufb a0, a2, a2; \
489	vpshufb a0, a3, a3; \
490	vpshufb a0, b0, b0; \
491	vpshufb a0, b1, b1; \
492	vpshufb a0, b2, b2; \
493	vpshufb a0, b3, b3; \
494	vpshufb a0, a1, a1; \
495	vpshufb a0, c0, c0; \
496	vpshufb a0, c1, c1; \
497	vpshufb a0, c2, c2; \
498	vpshufb a0, c3, c3; \
499	vpshufb a0, d0, d0; \
500	vpshufb a0, d1, d1; \
501	vpshufb a0, d2, d2; \
502	vpshufb a0, d3, d3; \
503	vmovdqu d3, st1; \
504	vmovdqu st0, d3; \
505	vpshufb a0, d3, a0; \
506	vmovdqu d2, st0; \
507	\
508	transpose_4x4(a0, b0, c0, d0, d2, d3); \
509	transpose_4x4(a1, b1, c1, d1, d2, d3); \
510	vmovdqu st0, d2; \
511	vmovdqu st1, d3; \
512	\
513	vmovdqu b0, st0; \
514	vmovdqu b1, st1; \
515	transpose_4x4(a2, b2, c2, d2, b0, b1); \
516	transpose_4x4(a3, b3, c3, d3, b0, b1); \
517	vmovdqu st0, b0; \
518	vmovdqu st1, b1; \
519	/* does not adjust output bytes inside vectors */
520
521/* load blocks to registers and apply pre-whitening */
522#define inpack32_pre(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
523		     y6, y7, rio, key) \
524	vpbroadcastq key, x0; \
525	vpshufb .Lpack_bswap, x0, x0; \
526	\
527	vpxor 0 * 32(rio), x0, y7; \
528	vpxor 1 * 32(rio), x0, y6; \
529	vpxor 2 * 32(rio), x0, y5; \
530	vpxor 3 * 32(rio), x0, y4; \
531	vpxor 4 * 32(rio), x0, y3; \
532	vpxor 5 * 32(rio), x0, y2; \
533	vpxor 6 * 32(rio), x0, y1; \
534	vpxor 7 * 32(rio), x0, y0; \
535	vpxor 8 * 32(rio), x0, x7; \
536	vpxor 9 * 32(rio), x0, x6; \
537	vpxor 10 * 32(rio), x0, x5; \
538	vpxor 11 * 32(rio), x0, x4; \
539	vpxor 12 * 32(rio), x0, x3; \
540	vpxor 13 * 32(rio), x0, x2; \
541	vpxor 14 * 32(rio), x0, x1; \
542	vpxor 15 * 32(rio), x0, x0;
543
544/* byteslice pre-whitened blocks and store to temporary memory */
545#define inpack32_post(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
546		      y6, y7, mem_ab, mem_cd) \
547	byteslice_16x16b_fast(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, \
548			      y4, y5, y6, y7, (mem_ab), (mem_cd)); \
549	\
550	vmovdqu x0, 0 * 32(mem_ab); \
551	vmovdqu x1, 1 * 32(mem_ab); \
552	vmovdqu x2, 2 * 32(mem_ab); \
553	vmovdqu x3, 3 * 32(mem_ab); \
554	vmovdqu x4, 4 * 32(mem_ab); \
555	vmovdqu x5, 5 * 32(mem_ab); \
556	vmovdqu x6, 6 * 32(mem_ab); \
557	vmovdqu x7, 7 * 32(mem_ab); \
558	vmovdqu y0, 0 * 32(mem_cd); \
559	vmovdqu y1, 1 * 32(mem_cd); \
560	vmovdqu y2, 2 * 32(mem_cd); \
561	vmovdqu y3, 3 * 32(mem_cd); \
562	vmovdqu y4, 4 * 32(mem_cd); \
563	vmovdqu y5, 5 * 32(mem_cd); \
564	vmovdqu y6, 6 * 32(mem_cd); \
565	vmovdqu y7, 7 * 32(mem_cd);
566
567/* de-byteslice, apply post-whitening and store blocks */
568#define outunpack32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, \
569		    y5, y6, y7, key, stack_tmp0, stack_tmp1) \
570	byteslice_16x16b_fast(y0, y4, x0, x4, y1, y5, x1, x5, y2, y6, x2, x6, \
571			      y3, y7, x3, x7, stack_tmp0, stack_tmp1); \
572	\
573	vmovdqu x0, stack_tmp0; \
574	\
575	vpbroadcastq key, x0; \
576	vpshufb .Lpack_bswap, x0, x0; \
577	\
578	vpxor x0, y7, y7; \
579	vpxor x0, y6, y6; \
580	vpxor x0, y5, y5; \
581	vpxor x0, y4, y4; \
582	vpxor x0, y3, y3; \
583	vpxor x0, y2, y2; \
584	vpxor x0, y1, y1; \
585	vpxor x0, y0, y0; \
586	vpxor x0, x7, x7; \
587	vpxor x0, x6, x6; \
588	vpxor x0, x5, x5; \
589	vpxor x0, x4, x4; \
590	vpxor x0, x3, x3; \
591	vpxor x0, x2, x2; \
592	vpxor x0, x1, x1; \
593	vpxor stack_tmp0, x0, x0;
594
595#define write_output(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
596		     y6, y7, rio) \
597	vmovdqu x0, 0 * 32(rio); \
598	vmovdqu x1, 1 * 32(rio); \
599	vmovdqu x2, 2 * 32(rio); \
600	vmovdqu x3, 3 * 32(rio); \
601	vmovdqu x4, 4 * 32(rio); \
602	vmovdqu x5, 5 * 32(rio); \
603	vmovdqu x6, 6 * 32(rio); \
604	vmovdqu x7, 7 * 32(rio); \
605	vmovdqu y0, 8 * 32(rio); \
606	vmovdqu y1, 9 * 32(rio); \
607	vmovdqu y2, 10 * 32(rio); \
608	vmovdqu y3, 11 * 32(rio); \
609	vmovdqu y4, 12 * 32(rio); \
610	vmovdqu y5, 13 * 32(rio); \
611	vmovdqu y6, 14 * 32(rio); \
612	vmovdqu y7, 15 * 32(rio);
613
614
615.section	.rodata.cst32.shufb_16x16b, "aM", @progbits, 32
616.align 32
617#define SHUFB_BYTES(idx) \
618	0 + (idx), 4 + (idx), 8 + (idx), 12 + (idx)
619.Lshufb_16x16b:
620	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
621	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
622
623.section	.rodata.cst32.pack_bswap, "aM", @progbits, 32
624.align 32
625.Lpack_bswap:
626	.long 0x00010203, 0x04050607, 0x80808080, 0x80808080
627	.long 0x00010203, 0x04050607, 0x80808080, 0x80808080
628
629/* NB: section is mergeable, all elements must be aligned 16-byte blocks */
630.section	.rodata.cst16, "aM", @progbits, 16
631.align 16
632
633/* For CTR-mode IV byteswap */
634.Lbswap128_mask:
635	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
636
637/* For XTS mode */
638.Lxts_gf128mul_and_shl1_mask_0:
639	.byte 0x87, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0
640.Lxts_gf128mul_and_shl1_mask_1:
641	.byte 0x0e, 1, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0
642
643/*
644 * pre-SubByte transform
645 *
646 * pre-lookup for sbox1, sbox2, sbox3:
647 *   swap_bitendianness(
648 *       isom_map_camellia_to_aes(
649 *           camellia_f(
650 *               swap_bitendianess(in)
651 *           )
652 *       )
653 *   )
654 *
655 * (note: '⊕ 0xc5' inside camellia_f())
656 */
657.Lpre_tf_lo_s1:
658	.byte 0x45, 0xe8, 0x40, 0xed, 0x2e, 0x83, 0x2b, 0x86
659	.byte 0x4b, 0xe6, 0x4e, 0xe3, 0x20, 0x8d, 0x25, 0x88
660.Lpre_tf_hi_s1:
661	.byte 0x00, 0x51, 0xf1, 0xa0, 0x8a, 0xdb, 0x7b, 0x2a
662	.byte 0x09, 0x58, 0xf8, 0xa9, 0x83, 0xd2, 0x72, 0x23
663
664/*
665 * pre-SubByte transform
666 *
667 * pre-lookup for sbox4:
668 *   swap_bitendianness(
669 *       isom_map_camellia_to_aes(
670 *           camellia_f(
671 *               swap_bitendianess(in <<< 1)
672 *           )
673 *       )
674 *   )
675 *
676 * (note: '⊕ 0xc5' inside camellia_f())
677 */
678.Lpre_tf_lo_s4:
679	.byte 0x45, 0x40, 0x2e, 0x2b, 0x4b, 0x4e, 0x20, 0x25
680	.byte 0x14, 0x11, 0x7f, 0x7a, 0x1a, 0x1f, 0x71, 0x74
681.Lpre_tf_hi_s4:
682	.byte 0x00, 0xf1, 0x8a, 0x7b, 0x09, 0xf8, 0x83, 0x72
683	.byte 0xad, 0x5c, 0x27, 0xd6, 0xa4, 0x55, 0x2e, 0xdf
684
685/*
686 * post-SubByte transform
687 *
688 * post-lookup for sbox1, sbox4:
689 *  swap_bitendianness(
690 *      camellia_h(
691 *          isom_map_aes_to_camellia(
692 *              swap_bitendianness(
693 *                  aes_inverse_affine_transform(in)
694 *              )
695 *          )
696 *      )
697 *  )
698 *
699 * (note: '⊕ 0x6e' inside camellia_h())
700 */
701.Lpost_tf_lo_s1:
702	.byte 0x3c, 0xcc, 0xcf, 0x3f, 0x32, 0xc2, 0xc1, 0x31
703	.byte 0xdc, 0x2c, 0x2f, 0xdf, 0xd2, 0x22, 0x21, 0xd1
704.Lpost_tf_hi_s1:
705	.byte 0x00, 0xf9, 0x86, 0x7f, 0xd7, 0x2e, 0x51, 0xa8
706	.byte 0xa4, 0x5d, 0x22, 0xdb, 0x73, 0x8a, 0xf5, 0x0c
707
708/*
709 * post-SubByte transform
710 *
711 * post-lookup for sbox2:
712 *  swap_bitendianness(
713 *      camellia_h(
714 *          isom_map_aes_to_camellia(
715 *              swap_bitendianness(
716 *                  aes_inverse_affine_transform(in)
717 *              )
718 *          )
719 *      )
720 *  ) <<< 1
721 *
722 * (note: '⊕ 0x6e' inside camellia_h())
723 */
724.Lpost_tf_lo_s2:
725	.byte 0x78, 0x99, 0x9f, 0x7e, 0x64, 0x85, 0x83, 0x62
726	.byte 0xb9, 0x58, 0x5e, 0xbf, 0xa5, 0x44, 0x42, 0xa3
727.Lpost_tf_hi_s2:
728	.byte 0x00, 0xf3, 0x0d, 0xfe, 0xaf, 0x5c, 0xa2, 0x51
729	.byte 0x49, 0xba, 0x44, 0xb7, 0xe6, 0x15, 0xeb, 0x18
730
731/*
732 * post-SubByte transform
733 *
734 * post-lookup for sbox3:
735 *  swap_bitendianness(
736 *      camellia_h(
737 *          isom_map_aes_to_camellia(
738 *              swap_bitendianness(
739 *                  aes_inverse_affine_transform(in)
740 *              )
741 *          )
742 *      )
743 *  ) >>> 1
744 *
745 * (note: '⊕ 0x6e' inside camellia_h())
746 */
747.Lpost_tf_lo_s3:
748	.byte 0x1e, 0x66, 0xe7, 0x9f, 0x19, 0x61, 0xe0, 0x98
749	.byte 0x6e, 0x16, 0x97, 0xef, 0x69, 0x11, 0x90, 0xe8
750.Lpost_tf_hi_s3:
751	.byte 0x00, 0xfc, 0x43, 0xbf, 0xeb, 0x17, 0xa8, 0x54
752	.byte 0x52, 0xae, 0x11, 0xed, 0xb9, 0x45, 0xfa, 0x06
753
754/* For isolating SubBytes from AESENCLAST, inverse shift row */
755.Linv_shift_row:
756	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
757	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
758
759.section	.rodata.cst4.L0f0f0f0f, "aM", @progbits, 4
760.align 4
761/* 4-bit mask */
762.L0f0f0f0f:
763	.long 0x0f0f0f0f
764
765.text
766
767.align 8
768__camellia_enc_blk32:
769	/* input:
770	 *	%rdi: ctx, CTX
771	 *	%rax: temporary storage, 512 bytes
772	 *	%ymm0..%ymm15: 32 plaintext blocks
773	 * output:
774	 *	%ymm0..%ymm15: 32 encrypted blocks, order swapped:
775	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
776	 */
777	FRAME_BEGIN
778
779	leaq 8 * 32(%rax), %rcx;
780
781	inpack32_post(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
782		      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
783		      %ymm15, %rax, %rcx);
784
785	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
786		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
787		     %ymm15, %rax, %rcx, 0);
788
789	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
790	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
791	      %ymm15,
792	      ((key_table + (8) * 8) + 0)(CTX),
793	      ((key_table + (8) * 8) + 4)(CTX),
794	      ((key_table + (8) * 8) + 8)(CTX),
795	      ((key_table + (8) * 8) + 12)(CTX));
796
797	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
798		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
799		     %ymm15, %rax, %rcx, 8);
800
801	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
802	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
803	      %ymm15,
804	      ((key_table + (16) * 8) + 0)(CTX),
805	      ((key_table + (16) * 8) + 4)(CTX),
806	      ((key_table + (16) * 8) + 8)(CTX),
807	      ((key_table + (16) * 8) + 12)(CTX));
808
809	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
810		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
811		     %ymm15, %rax, %rcx, 16);
812
813	movl $24, %r8d;
814	cmpl $16, key_length(CTX);
815	jne .Lenc_max32;
816
817.Lenc_done:
818	/* load CD for output */
819	vmovdqu 0 * 32(%rcx), %ymm8;
820	vmovdqu 1 * 32(%rcx), %ymm9;
821	vmovdqu 2 * 32(%rcx), %ymm10;
822	vmovdqu 3 * 32(%rcx), %ymm11;
823	vmovdqu 4 * 32(%rcx), %ymm12;
824	vmovdqu 5 * 32(%rcx), %ymm13;
825	vmovdqu 6 * 32(%rcx), %ymm14;
826	vmovdqu 7 * 32(%rcx), %ymm15;
827
828	outunpack32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
829		    %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
830		    %ymm15, (key_table)(CTX, %r8, 8), (%rax), 1 * 32(%rax));
831
832	FRAME_END
833	ret;
834
835.align 8
836.Lenc_max32:
837	movl $32, %r8d;
838
839	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
840	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
841	      %ymm15,
842	      ((key_table + (24) * 8) + 0)(CTX),
843	      ((key_table + (24) * 8) + 4)(CTX),
844	      ((key_table + (24) * 8) + 8)(CTX),
845	      ((key_table + (24) * 8) + 12)(CTX));
846
847	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
848		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
849		     %ymm15, %rax, %rcx, 24);
850
851	jmp .Lenc_done;
852ENDPROC(__camellia_enc_blk32)
853
854.align 8
855__camellia_dec_blk32:
856	/* input:
857	 *	%rdi: ctx, CTX
858	 *	%rax: temporary storage, 512 bytes
859	 *	%r8d: 24 for 16 byte key, 32 for larger
860	 *	%ymm0..%ymm15: 16 encrypted blocks
861	 * output:
862	 *	%ymm0..%ymm15: 16 plaintext blocks, order swapped:
863	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
864	 */
865	FRAME_BEGIN
866
867	leaq 8 * 32(%rax), %rcx;
868
869	inpack32_post(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
870		      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
871		      %ymm15, %rax, %rcx);
872
873	cmpl $32, %r8d;
874	je .Ldec_max32;
875
876.Ldec_max24:
877	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
878		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
879		     %ymm15, %rax, %rcx, 16);
880
881	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
882	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
883	      %ymm15,
884	      ((key_table + (16) * 8) + 8)(CTX),
885	      ((key_table + (16) * 8) + 12)(CTX),
886	      ((key_table + (16) * 8) + 0)(CTX),
887	      ((key_table + (16) * 8) + 4)(CTX));
888
889	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
890		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
891		     %ymm15, %rax, %rcx, 8);
892
893	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
894	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
895	      %ymm15,
896	      ((key_table + (8) * 8) + 8)(CTX),
897	      ((key_table + (8) * 8) + 12)(CTX),
898	      ((key_table + (8) * 8) + 0)(CTX),
899	      ((key_table + (8) * 8) + 4)(CTX));
900
901	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
902		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
903		     %ymm15, %rax, %rcx, 0);
904
905	/* load CD for output */
906	vmovdqu 0 * 32(%rcx), %ymm8;
907	vmovdqu 1 * 32(%rcx), %ymm9;
908	vmovdqu 2 * 32(%rcx), %ymm10;
909	vmovdqu 3 * 32(%rcx), %ymm11;
910	vmovdqu 4 * 32(%rcx), %ymm12;
911	vmovdqu 5 * 32(%rcx), %ymm13;
912	vmovdqu 6 * 32(%rcx), %ymm14;
913	vmovdqu 7 * 32(%rcx), %ymm15;
914
915	outunpack32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
916		    %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
917		    %ymm15, (key_table)(CTX), (%rax), 1 * 32(%rax));
918
919	FRAME_END
920	ret;
921
922.align 8
923.Ldec_max32:
924	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
925		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
926		     %ymm15, %rax, %rcx, 24);
927
928	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
929	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
930	      %ymm15,
931	      ((key_table + (24) * 8) + 8)(CTX),
932	      ((key_table + (24) * 8) + 12)(CTX),
933	      ((key_table + (24) * 8) + 0)(CTX),
934	      ((key_table + (24) * 8) + 4)(CTX));
935
936	jmp .Ldec_max24;
937ENDPROC(__camellia_dec_blk32)
938
939ENTRY(camellia_ecb_enc_32way)
940	/* input:
941	 *	%rdi: ctx, CTX
942	 *	%rsi: dst (32 blocks)
943	 *	%rdx: src (32 blocks)
944	 */
945	FRAME_BEGIN
946
947	vzeroupper;
948
949	inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
950		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
951		     %ymm15, %rdx, (key_table)(CTX));
952
953	/* now dst can be used as temporary buffer (even in src == dst case) */
954	movq	%rsi, %rax;
955
956	call __camellia_enc_blk32;
957
958	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
959		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
960		     %ymm8, %rsi);
961
962	vzeroupper;
963
964	FRAME_END
965	ret;
966ENDPROC(camellia_ecb_enc_32way)
967
968ENTRY(camellia_ecb_dec_32way)
969	/* input:
970	 *	%rdi: ctx, CTX
971	 *	%rsi: dst (32 blocks)
972	 *	%rdx: src (32 blocks)
973	 */
974	FRAME_BEGIN
975
976	vzeroupper;
977
978	cmpl $16, key_length(CTX);
979	movl $32, %r8d;
980	movl $24, %eax;
981	cmovel %eax, %r8d; /* max */
982
983	inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
984		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
985		     %ymm15, %rdx, (key_table)(CTX, %r8, 8));
986
987	/* now dst can be used as temporary buffer (even in src == dst case) */
988	movq	%rsi, %rax;
989
990	call __camellia_dec_blk32;
991
992	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
993		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
994		     %ymm8, %rsi);
995
996	vzeroupper;
997
998	FRAME_END
999	ret;
1000ENDPROC(camellia_ecb_dec_32way)
1001
1002ENTRY(camellia_cbc_dec_32way)
1003	/* input:
1004	 *	%rdi: ctx, CTX
1005	 *	%rsi: dst (32 blocks)
1006	 *	%rdx: src (32 blocks)
1007	 */
1008	FRAME_BEGIN
1009
1010	vzeroupper;
1011
1012	cmpl $16, key_length(CTX);
1013	movl $32, %r8d;
1014	movl $24, %eax;
1015	cmovel %eax, %r8d; /* max */
1016
1017	inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
1018		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
1019		     %ymm15, %rdx, (key_table)(CTX, %r8, 8));
1020
1021	movq %rsp, %r10;
1022	cmpq %rsi, %rdx;
1023	je .Lcbc_dec_use_stack;
1024
1025	/* dst can be used as temporary storage, src is not overwritten. */
1026	movq %rsi, %rax;
1027	jmp .Lcbc_dec_continue;
1028
1029.Lcbc_dec_use_stack:
1030	/*
1031	 * dst still in-use (because dst == src), so use stack for temporary
1032	 * storage.
1033	 */
1034	subq $(16 * 32), %rsp;
1035	movq %rsp, %rax;
1036
1037.Lcbc_dec_continue:
1038	call __camellia_dec_blk32;
1039
1040	vmovdqu %ymm7, (%rax);
1041	vpxor %ymm7, %ymm7, %ymm7;
1042	vinserti128 $1, (%rdx), %ymm7, %ymm7;
1043	vpxor (%rax), %ymm7, %ymm7;
1044	movq %r10, %rsp;
1045	vpxor (0 * 32 + 16)(%rdx), %ymm6, %ymm6;
1046	vpxor (1 * 32 + 16)(%rdx), %ymm5, %ymm5;
1047	vpxor (2 * 32 + 16)(%rdx), %ymm4, %ymm4;
1048	vpxor (3 * 32 + 16)(%rdx), %ymm3, %ymm3;
1049	vpxor (4 * 32 + 16)(%rdx), %ymm2, %ymm2;
1050	vpxor (5 * 32 + 16)(%rdx), %ymm1, %ymm1;
1051	vpxor (6 * 32 + 16)(%rdx), %ymm0, %ymm0;
1052	vpxor (7 * 32 + 16)(%rdx), %ymm15, %ymm15;
1053	vpxor (8 * 32 + 16)(%rdx), %ymm14, %ymm14;
1054	vpxor (9 * 32 + 16)(%rdx), %ymm13, %ymm13;
1055	vpxor (10 * 32 + 16)(%rdx), %ymm12, %ymm12;
1056	vpxor (11 * 32 + 16)(%rdx), %ymm11, %ymm11;
1057	vpxor (12 * 32 + 16)(%rdx), %ymm10, %ymm10;
1058	vpxor (13 * 32 + 16)(%rdx), %ymm9, %ymm9;
1059	vpxor (14 * 32 + 16)(%rdx), %ymm8, %ymm8;
1060	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
1061		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
1062		     %ymm8, %rsi);
1063
1064	vzeroupper;
1065
1066	FRAME_END
1067	ret;
1068ENDPROC(camellia_cbc_dec_32way)
1069
1070#define inc_le128(x, minus_one, tmp) \
1071	vpcmpeqq minus_one, x, tmp; \
1072	vpsubq minus_one, x, x; \
1073	vpslldq $8, tmp, tmp; \
1074	vpsubq tmp, x, x;
1075
1076#define add2_le128(x, minus_one, minus_two, tmp1, tmp2) \
1077	vpcmpeqq minus_one, x, tmp1; \
1078	vpcmpeqq minus_two, x, tmp2; \
1079	vpsubq minus_two, x, x; \
1080	vpor tmp2, tmp1, tmp1; \
1081	vpslldq $8, tmp1, tmp1; \
1082	vpsubq tmp1, x, x;
1083
1084ENTRY(camellia_ctr_32way)
1085	/* input:
1086	 *	%rdi: ctx, CTX
1087	 *	%rsi: dst (32 blocks)
1088	 *	%rdx: src (32 blocks)
1089	 *	%rcx: iv (little endian, 128bit)
1090	 */
1091	FRAME_BEGIN
1092
1093	vzeroupper;
1094
1095	movq %rsp, %r10;
1096	cmpq %rsi, %rdx;
1097	je .Lctr_use_stack;
1098
1099	/* dst can be used as temporary storage, src is not overwritten. */
1100	movq %rsi, %rax;
1101	jmp .Lctr_continue;
1102
1103.Lctr_use_stack:
1104	subq $(16 * 32), %rsp;
1105	movq %rsp, %rax;
1106
1107.Lctr_continue:
1108	vpcmpeqd %ymm15, %ymm15, %ymm15;
1109	vpsrldq $8, %ymm15, %ymm15; /* ab: -1:0 ; cd: -1:0 */
1110	vpaddq %ymm15, %ymm15, %ymm12; /* ab: -2:0 ; cd: -2:0 */
1111
1112	/* load IV and byteswap */
1113	vmovdqu (%rcx), %xmm0;
1114	vmovdqa %xmm0, %xmm1;
1115	inc_le128(%xmm0, %xmm15, %xmm14);
1116	vbroadcasti128 .Lbswap128_mask, %ymm14;
1117	vinserti128 $1, %xmm0, %ymm1, %ymm0;
1118	vpshufb %ymm14, %ymm0, %ymm13;
1119	vmovdqu %ymm13, 15 * 32(%rax);
1120
1121	/* construct IVs */
1122	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13); /* ab:le2 ; cd:le3 */
1123	vpshufb %ymm14, %ymm0, %ymm13;
1124	vmovdqu %ymm13, 14 * 32(%rax);
1125	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1126	vpshufb %ymm14, %ymm0, %ymm13;
1127	vmovdqu %ymm13, 13 * 32(%rax);
1128	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1129	vpshufb %ymm14, %ymm0, %ymm13;
1130	vmovdqu %ymm13, 12 * 32(%rax);
1131	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1132	vpshufb %ymm14, %ymm0, %ymm13;
1133	vmovdqu %ymm13, 11 * 32(%rax);
1134	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1135	vpshufb %ymm14, %ymm0, %ymm10;
1136	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1137	vpshufb %ymm14, %ymm0, %ymm9;
1138	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1139	vpshufb %ymm14, %ymm0, %ymm8;
1140	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1141	vpshufb %ymm14, %ymm0, %ymm7;
1142	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1143	vpshufb %ymm14, %ymm0, %ymm6;
1144	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1145	vpshufb %ymm14, %ymm0, %ymm5;
1146	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1147	vpshufb %ymm14, %ymm0, %ymm4;
1148	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1149	vpshufb %ymm14, %ymm0, %ymm3;
1150	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1151	vpshufb %ymm14, %ymm0, %ymm2;
1152	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1153	vpshufb %ymm14, %ymm0, %ymm1;
1154	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1155	vextracti128 $1, %ymm0, %xmm13;
1156	vpshufb %ymm14, %ymm0, %ymm0;
1157	inc_le128(%xmm13, %xmm15, %xmm14);
1158	vmovdqu %xmm13, (%rcx);
1159
1160	/* inpack32_pre: */
1161	vpbroadcastq (key_table)(CTX), %ymm15;
1162	vpshufb .Lpack_bswap, %ymm15, %ymm15;
1163	vpxor %ymm0, %ymm15, %ymm0;
1164	vpxor %ymm1, %ymm15, %ymm1;
1165	vpxor %ymm2, %ymm15, %ymm2;
1166	vpxor %ymm3, %ymm15, %ymm3;
1167	vpxor %ymm4, %ymm15, %ymm4;
1168	vpxor %ymm5, %ymm15, %ymm5;
1169	vpxor %ymm6, %ymm15, %ymm6;
1170	vpxor %ymm7, %ymm15, %ymm7;
1171	vpxor %ymm8, %ymm15, %ymm8;
1172	vpxor %ymm9, %ymm15, %ymm9;
1173	vpxor %ymm10, %ymm15, %ymm10;
1174	vpxor 11 * 32(%rax), %ymm15, %ymm11;
1175	vpxor 12 * 32(%rax), %ymm15, %ymm12;
1176	vpxor 13 * 32(%rax), %ymm15, %ymm13;
1177	vpxor 14 * 32(%rax), %ymm15, %ymm14;
1178	vpxor 15 * 32(%rax), %ymm15, %ymm15;
1179
1180	call __camellia_enc_blk32;
1181
1182	movq %r10, %rsp;
1183
1184	vpxor 0 * 32(%rdx), %ymm7, %ymm7;
1185	vpxor 1 * 32(%rdx), %ymm6, %ymm6;
1186	vpxor 2 * 32(%rdx), %ymm5, %ymm5;
1187	vpxor 3 * 32(%rdx), %ymm4, %ymm4;
1188	vpxor 4 * 32(%rdx), %ymm3, %ymm3;
1189	vpxor 5 * 32(%rdx), %ymm2, %ymm2;
1190	vpxor 6 * 32(%rdx), %ymm1, %ymm1;
1191	vpxor 7 * 32(%rdx), %ymm0, %ymm0;
1192	vpxor 8 * 32(%rdx), %ymm15, %ymm15;
1193	vpxor 9 * 32(%rdx), %ymm14, %ymm14;
1194	vpxor 10 * 32(%rdx), %ymm13, %ymm13;
1195	vpxor 11 * 32(%rdx), %ymm12, %ymm12;
1196	vpxor 12 * 32(%rdx), %ymm11, %ymm11;
1197	vpxor 13 * 32(%rdx), %ymm10, %ymm10;
1198	vpxor 14 * 32(%rdx), %ymm9, %ymm9;
1199	vpxor 15 * 32(%rdx), %ymm8, %ymm8;
1200	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
1201		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
1202		     %ymm8, %rsi);
1203
1204	vzeroupper;
1205
1206	FRAME_END
1207	ret;
1208ENDPROC(camellia_ctr_32way)
1209
1210#define gf128mul_x_ble(iv, mask, tmp) \
1211	vpsrad $31, iv, tmp; \
1212	vpaddq iv, iv, iv; \
1213	vpshufd $0x13, tmp, tmp; \
1214	vpand mask, tmp, tmp; \
1215	vpxor tmp, iv, iv;
1216
1217#define gf128mul_x2_ble(iv, mask1, mask2, tmp0, tmp1) \
1218	vpsrad $31, iv, tmp0; \
1219	vpaddq iv, iv, tmp1; \
1220	vpsllq $2, iv, iv; \
1221	vpshufd $0x13, tmp0, tmp0; \
1222	vpsrad $31, tmp1, tmp1; \
1223	vpand mask2, tmp0, tmp0; \
1224	vpshufd $0x13, tmp1, tmp1; \
1225	vpxor tmp0, iv, iv; \
1226	vpand mask1, tmp1, tmp1; \
1227	vpxor tmp1, iv, iv;
1228
1229.align 8
1230camellia_xts_crypt_32way:
1231	/* input:
1232	 *	%rdi: ctx, CTX
1233	 *	%rsi: dst (32 blocks)
1234	 *	%rdx: src (32 blocks)
1235	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
1236	 *	%r8: index for input whitening key
1237	 *	%r9: pointer to  __camellia_enc_blk32 or __camellia_dec_blk32
1238	 */
1239	FRAME_BEGIN
1240
1241	vzeroupper;
1242
1243	subq $(16 * 32), %rsp;
1244	movq %rsp, %rax;
1245
1246	vbroadcasti128 .Lxts_gf128mul_and_shl1_mask_0, %ymm12;
1247
1248	/* load IV and construct second IV */
1249	vmovdqu (%rcx), %xmm0;
1250	vmovdqa %xmm0, %xmm15;
1251	gf128mul_x_ble(%xmm0, %xmm12, %xmm13);
1252	vbroadcasti128 .Lxts_gf128mul_and_shl1_mask_1, %ymm13;
1253	vinserti128 $1, %xmm0, %ymm15, %ymm0;
1254	vpxor 0 * 32(%rdx), %ymm0, %ymm15;
1255	vmovdqu %ymm15, 15 * 32(%rax);
1256	vmovdqu %ymm0, 0 * 32(%rsi);
1257
1258	/* construct IVs */
1259	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1260	vpxor 1 * 32(%rdx), %ymm0, %ymm15;
1261	vmovdqu %ymm15, 14 * 32(%rax);
1262	vmovdqu %ymm0, 1 * 32(%rsi);
1263
1264	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1265	vpxor 2 * 32(%rdx), %ymm0, %ymm15;
1266	vmovdqu %ymm15, 13 * 32(%rax);
1267	vmovdqu %ymm0, 2 * 32(%rsi);
1268
1269	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1270	vpxor 3 * 32(%rdx), %ymm0, %ymm15;
1271	vmovdqu %ymm15, 12 * 32(%rax);
1272	vmovdqu %ymm0, 3 * 32(%rsi);
1273
1274	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1275	vpxor 4 * 32(%rdx), %ymm0, %ymm11;
1276	vmovdqu %ymm0, 4 * 32(%rsi);
1277
1278	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1279	vpxor 5 * 32(%rdx), %ymm0, %ymm10;
1280	vmovdqu %ymm0, 5 * 32(%rsi);
1281
1282	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1283	vpxor 6 * 32(%rdx), %ymm0, %ymm9;
1284	vmovdqu %ymm0, 6 * 32(%rsi);
1285
1286	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1287	vpxor 7 * 32(%rdx), %ymm0, %ymm8;
1288	vmovdqu %ymm0, 7 * 32(%rsi);
1289
1290	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1291	vpxor 8 * 32(%rdx), %ymm0, %ymm7;
1292	vmovdqu %ymm0, 8 * 32(%rsi);
1293
1294	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1295	vpxor 9 * 32(%rdx), %ymm0, %ymm6;
1296	vmovdqu %ymm0, 9 * 32(%rsi);
1297
1298	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1299	vpxor 10 * 32(%rdx), %ymm0, %ymm5;
1300	vmovdqu %ymm0, 10 * 32(%rsi);
1301
1302	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1303	vpxor 11 * 32(%rdx), %ymm0, %ymm4;
1304	vmovdqu %ymm0, 11 * 32(%rsi);
1305
1306	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1307	vpxor 12 * 32(%rdx), %ymm0, %ymm3;
1308	vmovdqu %ymm0, 12 * 32(%rsi);
1309
1310	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1311	vpxor 13 * 32(%rdx), %ymm0, %ymm2;
1312	vmovdqu %ymm0, 13 * 32(%rsi);
1313
1314	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1315	vpxor 14 * 32(%rdx), %ymm0, %ymm1;
1316	vmovdqu %ymm0, 14 * 32(%rsi);
1317
1318	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1319	vpxor 15 * 32(%rdx), %ymm0, %ymm15;
1320	vmovdqu %ymm15, 0 * 32(%rax);
1321	vmovdqu %ymm0, 15 * 32(%rsi);
1322
1323	vextracti128 $1, %ymm0, %xmm0;
1324	gf128mul_x_ble(%xmm0, %xmm12, %xmm15);
1325	vmovdqu %xmm0, (%rcx);
1326
1327	/* inpack32_pre: */
1328	vpbroadcastq (key_table)(CTX, %r8, 8), %ymm15;
1329	vpshufb .Lpack_bswap, %ymm15, %ymm15;
1330	vpxor 0 * 32(%rax), %ymm15, %ymm0;
1331	vpxor %ymm1, %ymm15, %ymm1;
1332	vpxor %ymm2, %ymm15, %ymm2;
1333	vpxor %ymm3, %ymm15, %ymm3;
1334	vpxor %ymm4, %ymm15, %ymm4;
1335	vpxor %ymm5, %ymm15, %ymm5;
1336	vpxor %ymm6, %ymm15, %ymm6;
1337	vpxor %ymm7, %ymm15, %ymm7;
1338	vpxor %ymm8, %ymm15, %ymm8;
1339	vpxor %ymm9, %ymm15, %ymm9;
1340	vpxor %ymm10, %ymm15, %ymm10;
1341	vpxor %ymm11, %ymm15, %ymm11;
1342	vpxor 12 * 32(%rax), %ymm15, %ymm12;
1343	vpxor 13 * 32(%rax), %ymm15, %ymm13;
1344	vpxor 14 * 32(%rax), %ymm15, %ymm14;
1345	vpxor 15 * 32(%rax), %ymm15, %ymm15;
1346
1347	CALL_NOSPEC %r9;
1348
1349	addq $(16 * 32), %rsp;
1350
1351	vpxor 0 * 32(%rsi), %ymm7, %ymm7;
1352	vpxor 1 * 32(%rsi), %ymm6, %ymm6;
1353	vpxor 2 * 32(%rsi), %ymm5, %ymm5;
1354	vpxor 3 * 32(%rsi), %ymm4, %ymm4;
1355	vpxor 4 * 32(%rsi), %ymm3, %ymm3;
1356	vpxor 5 * 32(%rsi), %ymm2, %ymm2;
1357	vpxor 6 * 32(%rsi), %ymm1, %ymm1;
1358	vpxor 7 * 32(%rsi), %ymm0, %ymm0;
1359	vpxor 8 * 32(%rsi), %ymm15, %ymm15;
1360	vpxor 9 * 32(%rsi), %ymm14, %ymm14;
1361	vpxor 10 * 32(%rsi), %ymm13, %ymm13;
1362	vpxor 11 * 32(%rsi), %ymm12, %ymm12;
1363	vpxor 12 * 32(%rsi), %ymm11, %ymm11;
1364	vpxor 13 * 32(%rsi), %ymm10, %ymm10;
1365	vpxor 14 * 32(%rsi), %ymm9, %ymm9;
1366	vpxor 15 * 32(%rsi), %ymm8, %ymm8;
1367	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
1368		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
1369		     %ymm8, %rsi);
1370
1371	vzeroupper;
1372
1373	FRAME_END
1374	ret;
1375ENDPROC(camellia_xts_crypt_32way)
1376
1377ENTRY(camellia_xts_enc_32way)
1378	/* input:
1379	 *	%rdi: ctx, CTX
1380	 *	%rsi: dst (32 blocks)
1381	 *	%rdx: src (32 blocks)
1382	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
1383	 */
1384
1385	xorl %r8d, %r8d; /* input whitening key, 0 for enc */
1386
1387	leaq __camellia_enc_blk32, %r9;
1388
1389	jmp camellia_xts_crypt_32way;
1390ENDPROC(camellia_xts_enc_32way)
1391
1392ENTRY(camellia_xts_dec_32way)
1393	/* input:
1394	 *	%rdi: ctx, CTX
1395	 *	%rsi: dst (32 blocks)
1396	 *	%rdx: src (32 blocks)
1397	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
1398	 */
1399
1400	cmpl $16, key_length(CTX);
1401	movl $32, %r8d;
1402	movl $24, %eax;
1403	cmovel %eax, %r8d;  /* input whitening key, last for dec */
1404
1405	leaq __camellia_dec_blk32, %r9;
1406
1407	jmp camellia_xts_crypt_32way;
1408ENDPROC(camellia_xts_dec_32way)
1409