1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * x86_64/AVX2/AES-NI assembler implementation of Camellia
4 *
5 * Copyright © 2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
6 */
7
8#include <linux/linkage.h>
9#include <asm/frame.h>
10#include <asm/nospec-branch.h>
11
12#define CAMELLIA_TABLE_BYTE_LEN 272
13
14/* struct camellia_ctx: */
15#define key_table 0
16#define key_length CAMELLIA_TABLE_BYTE_LEN
17
18/* register macros */
19#define CTX %rdi
20#define RIO %r8
21
22/**********************************************************************
23  helper macros
24 **********************************************************************/
25#define filter_8bit(x, lo_t, hi_t, mask4bit, tmp0) \
26	vpand x, mask4bit, tmp0; \
27	vpandn x, mask4bit, x; \
28	vpsrld $4, x, x; \
29	\
30	vpshufb tmp0, lo_t, tmp0; \
31	vpshufb x, hi_t, x; \
32	vpxor tmp0, x, x;
33
34#define ymm0_x xmm0
35#define ymm1_x xmm1
36#define ymm2_x xmm2
37#define ymm3_x xmm3
38#define ymm4_x xmm4
39#define ymm5_x xmm5
40#define ymm6_x xmm6
41#define ymm7_x xmm7
42#define ymm8_x xmm8
43#define ymm9_x xmm9
44#define ymm10_x xmm10
45#define ymm11_x xmm11
46#define ymm12_x xmm12
47#define ymm13_x xmm13
48#define ymm14_x xmm14
49#define ymm15_x xmm15
50
51/**********************************************************************
52  32-way camellia
53 **********************************************************************/
54
55/*
56 * IN:
57 *   x0..x7: byte-sliced AB state
58 *   mem_cd: register pointer storing CD state
59 *   key: index for key material
60 * OUT:
61 *   x0..x7: new byte-sliced CD state
62 */
63#define roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, t0, t1, t2, t3, t4, t5, t6, \
64		  t7, mem_cd, key) \
65	/* \
66	 * S-function with AES subbytes \
67	 */ \
68	vbroadcasti128 .Linv_shift_row, t4; \
69	vpbroadcastd .L0f0f0f0f, t7; \
70	vbroadcasti128 .Lpre_tf_lo_s1, t5; \
71	vbroadcasti128 .Lpre_tf_hi_s1, t6; \
72	vbroadcasti128 .Lpre_tf_lo_s4, t2; \
73	vbroadcasti128 .Lpre_tf_hi_s4, t3; \
74	\
75	/* AES inverse shift rows */ \
76	vpshufb t4, x0, x0; \
77	vpshufb t4, x7, x7; \
78	vpshufb t4, x3, x3; \
79	vpshufb t4, x6, x6; \
80	vpshufb t4, x2, x2; \
81	vpshufb t4, x5, x5; \
82	vpshufb t4, x1, x1; \
83	vpshufb t4, x4, x4; \
84	\
85	/* prefilter sboxes 1, 2 and 3 */ \
86	/* prefilter sbox 4 */ \
87	filter_8bit(x0, t5, t6, t7, t4); \
88	filter_8bit(x7, t5, t6, t7, t4); \
89	vextracti128 $1, x0, t0##_x; \
90	vextracti128 $1, x7, t1##_x; \
91	filter_8bit(x3, t2, t3, t7, t4); \
92	filter_8bit(x6, t2, t3, t7, t4); \
93	vextracti128 $1, x3, t3##_x; \
94	vextracti128 $1, x6, t2##_x; \
95	filter_8bit(x2, t5, t6, t7, t4); \
96	filter_8bit(x5, t5, t6, t7, t4); \
97	filter_8bit(x1, t5, t6, t7, t4); \
98	filter_8bit(x4, t5, t6, t7, t4); \
99	\
100	vpxor t4##_x, t4##_x, t4##_x; \
101	\
102	/* AES subbytes + AES shift rows */ \
103	vextracti128 $1, x2, t6##_x; \
104	vextracti128 $1, x5, t5##_x; \
105	vaesenclast t4##_x, x0##_x, x0##_x; \
106	vaesenclast t4##_x, t0##_x, t0##_x; \
107	vinserti128 $1, t0##_x, x0, x0; \
108	vaesenclast t4##_x, x7##_x, x7##_x; \
109	vaesenclast t4##_x, t1##_x, t1##_x; \
110	vinserti128 $1, t1##_x, x7, x7; \
111	vaesenclast t4##_x, x3##_x, x3##_x; \
112	vaesenclast t4##_x, t3##_x, t3##_x; \
113	vinserti128 $1, t3##_x, x3, x3; \
114	vaesenclast t4##_x, x6##_x, x6##_x; \
115	vaesenclast t4##_x, t2##_x, t2##_x; \
116	vinserti128 $1, t2##_x, x6, x6; \
117	vextracti128 $1, x1, t3##_x; \
118	vextracti128 $1, x4, t2##_x; \
119	vbroadcasti128 .Lpost_tf_lo_s1, t0; \
120	vbroadcasti128 .Lpost_tf_hi_s1, t1; \
121	vaesenclast t4##_x, x2##_x, x2##_x; \
122	vaesenclast t4##_x, t6##_x, t6##_x; \
123	vinserti128 $1, t6##_x, x2, x2; \
124	vaesenclast t4##_x, x5##_x, x5##_x; \
125	vaesenclast t4##_x, t5##_x, t5##_x; \
126	vinserti128 $1, t5##_x, x5, x5; \
127	vaesenclast t4##_x, x1##_x, x1##_x; \
128	vaesenclast t4##_x, t3##_x, t3##_x; \
129	vinserti128 $1, t3##_x, x1, x1; \
130	vaesenclast t4##_x, x4##_x, x4##_x; \
131	vaesenclast t4##_x, t2##_x, t2##_x; \
132	vinserti128 $1, t2##_x, x4, x4; \
133	\
134	/* postfilter sboxes 1 and 4 */ \
135	vbroadcasti128 .Lpost_tf_lo_s3, t2; \
136	vbroadcasti128 .Lpost_tf_hi_s3, t3; \
137	filter_8bit(x0, t0, t1, t7, t6); \
138	filter_8bit(x7, t0, t1, t7, t6); \
139	filter_8bit(x3, t0, t1, t7, t6); \
140	filter_8bit(x6, t0, t1, t7, t6); \
141	\
142	/* postfilter sbox 3 */ \
143	vbroadcasti128 .Lpost_tf_lo_s2, t4; \
144	vbroadcasti128 .Lpost_tf_hi_s2, t5; \
145	filter_8bit(x2, t2, t3, t7, t6); \
146	filter_8bit(x5, t2, t3, t7, t6); \
147	\
148	vpbroadcastq key, t0; /* higher 64-bit duplicate ignored */ \
149	\
150	/* postfilter sbox 2 */ \
151	filter_8bit(x1, t4, t5, t7, t2); \
152	filter_8bit(x4, t4, t5, t7, t2); \
153	vpxor t7, t7, t7; \
154	\
155	vpsrldq $1, t0, t1; \
156	vpsrldq $2, t0, t2; \
157	vpshufb t7, t1, t1; \
158	vpsrldq $3, t0, t3; \
159	\
160	/* P-function */ \
161	vpxor x5, x0, x0; \
162	vpxor x6, x1, x1; \
163	vpxor x7, x2, x2; \
164	vpxor x4, x3, x3; \
165	\
166	vpshufb t7, t2, t2; \
167	vpsrldq $4, t0, t4; \
168	vpshufb t7, t3, t3; \
169	vpsrldq $5, t0, t5; \
170	vpshufb t7, t4, t4; \
171	\
172	vpxor x2, x4, x4; \
173	vpxor x3, x5, x5; \
174	vpxor x0, x6, x6; \
175	vpxor x1, x7, x7; \
176	\
177	vpsrldq $6, t0, t6; \
178	vpshufb t7, t5, t5; \
179	vpshufb t7, t6, t6; \
180	\
181	vpxor x7, x0, x0; \
182	vpxor x4, x1, x1; \
183	vpxor x5, x2, x2; \
184	vpxor x6, x3, x3; \
185	\
186	vpxor x3, x4, x4; \
187	vpxor x0, x5, x5; \
188	vpxor x1, x6, x6; \
189	vpxor x2, x7, x7; /* note: high and low parts swapped */ \
190	\
191	/* Add key material and result to CD (x becomes new CD) */ \
192	\
193	vpxor t6, x1, x1; \
194	vpxor 5 * 32(mem_cd), x1, x1; \
195	\
196	vpsrldq $7, t0, t6; \
197	vpshufb t7, t0, t0; \
198	vpshufb t7, t6, t7; \
199	\
200	vpxor t7, x0, x0; \
201	vpxor 4 * 32(mem_cd), x0, x0; \
202	\
203	vpxor t5, x2, x2; \
204	vpxor 6 * 32(mem_cd), x2, x2; \
205	\
206	vpxor t4, x3, x3; \
207	vpxor 7 * 32(mem_cd), x3, x3; \
208	\
209	vpxor t3, x4, x4; \
210	vpxor 0 * 32(mem_cd), x4, x4; \
211	\
212	vpxor t2, x5, x5; \
213	vpxor 1 * 32(mem_cd), x5, x5; \
214	\
215	vpxor t1, x6, x6; \
216	vpxor 2 * 32(mem_cd), x6, x6; \
217	\
218	vpxor t0, x7, x7; \
219	vpxor 3 * 32(mem_cd), x7, x7;
220
221/*
222 * Size optimization... with inlined roundsm32 binary would be over 5 times
223 * larger and would only marginally faster.
224 */
225.align 8
226SYM_FUNC_START_LOCAL(roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
227	roundsm32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
228		  %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14, %ymm15,
229		  %rcx, (%r9));
230	ret;
231SYM_FUNC_END(roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
232
233.align 8
234SYM_FUNC_START_LOCAL(roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
235	roundsm32(%ymm4, %ymm5, %ymm6, %ymm7, %ymm0, %ymm1, %ymm2, %ymm3,
236		  %ymm12, %ymm13, %ymm14, %ymm15, %ymm8, %ymm9, %ymm10, %ymm11,
237		  %rax, (%r9));
238	ret;
239SYM_FUNC_END(roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
240
241/*
242 * IN/OUT:
243 *  x0..x7: byte-sliced AB state preloaded
244 *  mem_ab: byte-sliced AB state in memory
245 *  mem_cb: byte-sliced CD state in memory
246 */
247#define two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
248		      y6, y7, mem_ab, mem_cd, i, dir, store_ab) \
249	leaq (key_table + (i) * 8)(CTX), %r9; \
250	call roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd; \
251	\
252	vmovdqu x0, 4 * 32(mem_cd); \
253	vmovdqu x1, 5 * 32(mem_cd); \
254	vmovdqu x2, 6 * 32(mem_cd); \
255	vmovdqu x3, 7 * 32(mem_cd); \
256	vmovdqu x4, 0 * 32(mem_cd); \
257	vmovdqu x5, 1 * 32(mem_cd); \
258	vmovdqu x6, 2 * 32(mem_cd); \
259	vmovdqu x7, 3 * 32(mem_cd); \
260	\
261	leaq (key_table + ((i) + (dir)) * 8)(CTX), %r9; \
262	call roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab; \
263	\
264	store_ab(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab);
265
266#define dummy_store(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) /* do nothing */
267
268#define store_ab_state(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) \
269	/* Store new AB state */ \
270	vmovdqu x4, 4 * 32(mem_ab); \
271	vmovdqu x5, 5 * 32(mem_ab); \
272	vmovdqu x6, 6 * 32(mem_ab); \
273	vmovdqu x7, 7 * 32(mem_ab); \
274	vmovdqu x0, 0 * 32(mem_ab); \
275	vmovdqu x1, 1 * 32(mem_ab); \
276	vmovdqu x2, 2 * 32(mem_ab); \
277	vmovdqu x3, 3 * 32(mem_ab);
278
279#define enc_rounds32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
280		      y6, y7, mem_ab, mem_cd, i) \
281	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
282		      y6, y7, mem_ab, mem_cd, (i) + 2, 1, store_ab_state); \
283	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
284		      y6, y7, mem_ab, mem_cd, (i) + 4, 1, store_ab_state); \
285	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
286		      y6, y7, mem_ab, mem_cd, (i) + 6, 1, dummy_store);
287
288#define dec_rounds32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
289		      y6, y7, mem_ab, mem_cd, i) \
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) + 7, -1, store_ab_state); \
292	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
293		      y6, y7, mem_ab, mem_cd, (i) + 5, -1, store_ab_state); \
294	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
295		      y6, y7, mem_ab, mem_cd, (i) + 3, -1, dummy_store);
296
297/*
298 * IN:
299 *  v0..3: byte-sliced 32-bit integers
300 * OUT:
301 *  v0..3: (IN <<< 1)
302 */
303#define rol32_1_32(v0, v1, v2, v3, t0, t1, t2, zero) \
304	vpcmpgtb v0, zero, t0; \
305	vpaddb v0, v0, v0; \
306	vpabsb t0, t0; \
307	\
308	vpcmpgtb v1, zero, t1; \
309	vpaddb v1, v1, v1; \
310	vpabsb t1, t1; \
311	\
312	vpcmpgtb v2, zero, t2; \
313	vpaddb v2, v2, v2; \
314	vpabsb t2, t2; \
315	\
316	vpor t0, v1, v1; \
317	\
318	vpcmpgtb v3, zero, t0; \
319	vpaddb v3, v3, v3; \
320	vpabsb t0, t0; \
321	\
322	vpor t1, v2, v2; \
323	vpor t2, v3, v3; \
324	vpor t0, v0, v0;
325
326/*
327 * IN:
328 *   r: byte-sliced AB state in memory
329 *   l: byte-sliced CD state in memory
330 * OUT:
331 *   x0..x7: new byte-sliced CD state
332 */
333#define fls32(l, l0, l1, l2, l3, l4, l5, l6, l7, r, t0, t1, t2, t3, tt0, \
334	      tt1, tt2, tt3, kll, klr, krl, krr) \
335	/* \
336	 * t0 = kll; \
337	 * t0 &= ll; \
338	 * lr ^= rol32(t0, 1); \
339	 */ \
340	vpbroadcastd kll, t0; /* only lowest 32-bit used */ \
341	vpxor tt0, tt0, tt0; \
342	vpshufb tt0, t0, t3; \
343	vpsrldq $1, t0, t0; \
344	vpshufb tt0, t0, t2; \
345	vpsrldq $1, t0, t0; \
346	vpshufb tt0, t0, t1; \
347	vpsrldq $1, t0, t0; \
348	vpshufb tt0, t0, t0; \
349	\
350	vpand l0, t0, t0; \
351	vpand l1, t1, t1; \
352	vpand l2, t2, t2; \
353	vpand l3, t3, t3; \
354	\
355	rol32_1_32(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
356	\
357	vpxor l4, t0, l4; \
358	vpbroadcastd krr, t0; /* only lowest 32-bit used */ \
359	vmovdqu l4, 4 * 32(l); \
360	vpxor l5, t1, l5; \
361	vmovdqu l5, 5 * 32(l); \
362	vpxor l6, t2, l6; \
363	vmovdqu l6, 6 * 32(l); \
364	vpxor l7, t3, l7; \
365	vmovdqu l7, 7 * 32(l); \
366	\
367	/* \
368	 * t2 = krr; \
369	 * t2 |= rr; \
370	 * rl ^= t2; \
371	 */ \
372	\
373	vpshufb tt0, t0, t3; \
374	vpsrldq $1, t0, t0; \
375	vpshufb tt0, t0, t2; \
376	vpsrldq $1, t0, t0; \
377	vpshufb tt0, t0, t1; \
378	vpsrldq $1, t0, t0; \
379	vpshufb tt0, t0, t0; \
380	\
381	vpor 4 * 32(r), t0, t0; \
382	vpor 5 * 32(r), t1, t1; \
383	vpor 6 * 32(r), t2, t2; \
384	vpor 7 * 32(r), t3, t3; \
385	\
386	vpxor 0 * 32(r), t0, t0; \
387	vpxor 1 * 32(r), t1, t1; \
388	vpxor 2 * 32(r), t2, t2; \
389	vpxor 3 * 32(r), t3, t3; \
390	vmovdqu t0, 0 * 32(r); \
391	vpbroadcastd krl, t0; /* only lowest 32-bit used */ \
392	vmovdqu t1, 1 * 32(r); \
393	vmovdqu t2, 2 * 32(r); \
394	vmovdqu t3, 3 * 32(r); \
395	\
396	/* \
397	 * t2 = krl; \
398	 * t2 &= rl; \
399	 * rr ^= rol32(t2, 1); \
400	 */ \
401	vpshufb tt0, t0, t3; \
402	vpsrldq $1, t0, t0; \
403	vpshufb tt0, t0, t2; \
404	vpsrldq $1, t0, t0; \
405	vpshufb tt0, t0, t1; \
406	vpsrldq $1, t0, t0; \
407	vpshufb tt0, t0, t0; \
408	\
409	vpand 0 * 32(r), t0, t0; \
410	vpand 1 * 32(r), t1, t1; \
411	vpand 2 * 32(r), t2, t2; \
412	vpand 3 * 32(r), t3, t3; \
413	\
414	rol32_1_32(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
415	\
416	vpxor 4 * 32(r), t0, t0; \
417	vpxor 5 * 32(r), t1, t1; \
418	vpxor 6 * 32(r), t2, t2; \
419	vpxor 7 * 32(r), t3, t3; \
420	vmovdqu t0, 4 * 32(r); \
421	vpbroadcastd klr, t0; /* only lowest 32-bit used */ \
422	vmovdqu t1, 5 * 32(r); \
423	vmovdqu t2, 6 * 32(r); \
424	vmovdqu t3, 7 * 32(r); \
425	\
426	/* \
427	 * t0 = klr; \
428	 * t0 |= lr; \
429	 * ll ^= t0; \
430	 */ \
431	\
432	vpshufb tt0, t0, t3; \
433	vpsrldq $1, t0, t0; \
434	vpshufb tt0, t0, t2; \
435	vpsrldq $1, t0, t0; \
436	vpshufb tt0, t0, t1; \
437	vpsrldq $1, t0, t0; \
438	vpshufb tt0, t0, t0; \
439	\
440	vpor l4, t0, t0; \
441	vpor l5, t1, t1; \
442	vpor l6, t2, t2; \
443	vpor l7, t3, t3; \
444	\
445	vpxor l0, t0, l0; \
446	vmovdqu l0, 0 * 32(l); \
447	vpxor l1, t1, l1; \
448	vmovdqu l1, 1 * 32(l); \
449	vpxor l2, t2, l2; \
450	vmovdqu l2, 2 * 32(l); \
451	vpxor l3, t3, l3; \
452	vmovdqu l3, 3 * 32(l);
453
454#define transpose_4x4(x0, x1, x2, x3, t1, t2) \
455	vpunpckhdq x1, x0, t2; \
456	vpunpckldq x1, x0, x0; \
457	\
458	vpunpckldq x3, x2, t1; \
459	vpunpckhdq x3, x2, x2; \
460	\
461	vpunpckhqdq t1, x0, x1; \
462	vpunpcklqdq t1, x0, x0; \
463	\
464	vpunpckhqdq x2, t2, x3; \
465	vpunpcklqdq x2, t2, x2;
466
467#define byteslice_16x16b_fast(a0, b0, c0, d0, a1, b1, c1, d1, a2, b2, c2, d2, \
468			      a3, b3, c3, d3, st0, st1) \
469	vmovdqu d2, st0; \
470	vmovdqu d3, st1; \
471	transpose_4x4(a0, a1, a2, a3, d2, d3); \
472	transpose_4x4(b0, b1, b2, b3, d2, d3); \
473	vmovdqu st0, d2; \
474	vmovdqu st1, d3; \
475	\
476	vmovdqu a0, st0; \
477	vmovdqu a1, st1; \
478	transpose_4x4(c0, c1, c2, c3, a0, a1); \
479	transpose_4x4(d0, d1, d2, d3, a0, a1); \
480	\
481	vbroadcasti128 .Lshufb_16x16b, a0; \
482	vmovdqu st1, a1; \
483	vpshufb a0, a2, a2; \
484	vpshufb a0, a3, a3; \
485	vpshufb a0, b0, b0; \
486	vpshufb a0, b1, b1; \
487	vpshufb a0, b2, b2; \
488	vpshufb a0, b3, b3; \
489	vpshufb a0, a1, a1; \
490	vpshufb a0, c0, c0; \
491	vpshufb a0, c1, c1; \
492	vpshufb a0, c2, c2; \
493	vpshufb a0, c3, c3; \
494	vpshufb a0, d0, d0; \
495	vpshufb a0, d1, d1; \
496	vpshufb a0, d2, d2; \
497	vpshufb a0, d3, d3; \
498	vmovdqu d3, st1; \
499	vmovdqu st0, d3; \
500	vpshufb a0, d3, a0; \
501	vmovdqu d2, st0; \
502	\
503	transpose_4x4(a0, b0, c0, d0, d2, d3); \
504	transpose_4x4(a1, b1, c1, d1, d2, d3); \
505	vmovdqu st0, d2; \
506	vmovdqu st1, d3; \
507	\
508	vmovdqu b0, st0; \
509	vmovdqu b1, st1; \
510	transpose_4x4(a2, b2, c2, d2, b0, b1); \
511	transpose_4x4(a3, b3, c3, d3, b0, b1); \
512	vmovdqu st0, b0; \
513	vmovdqu st1, b1; \
514	/* does not adjust output bytes inside vectors */
515
516/* load blocks to registers and apply pre-whitening */
517#define inpack32_pre(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
518		     y6, y7, rio, key) \
519	vpbroadcastq key, x0; \
520	vpshufb .Lpack_bswap, x0, x0; \
521	\
522	vpxor 0 * 32(rio), x0, y7; \
523	vpxor 1 * 32(rio), x0, y6; \
524	vpxor 2 * 32(rio), x0, y5; \
525	vpxor 3 * 32(rio), x0, y4; \
526	vpxor 4 * 32(rio), x0, y3; \
527	vpxor 5 * 32(rio), x0, y2; \
528	vpxor 6 * 32(rio), x0, y1; \
529	vpxor 7 * 32(rio), x0, y0; \
530	vpxor 8 * 32(rio), x0, x7; \
531	vpxor 9 * 32(rio), x0, x6; \
532	vpxor 10 * 32(rio), x0, x5; \
533	vpxor 11 * 32(rio), x0, x4; \
534	vpxor 12 * 32(rio), x0, x3; \
535	vpxor 13 * 32(rio), x0, x2; \
536	vpxor 14 * 32(rio), x0, x1; \
537	vpxor 15 * 32(rio), x0, x0;
538
539/* byteslice pre-whitened blocks and store to temporary memory */
540#define inpack32_post(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
541		      y6, y7, mem_ab, mem_cd) \
542	byteslice_16x16b_fast(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, \
543			      y4, y5, y6, y7, (mem_ab), (mem_cd)); \
544	\
545	vmovdqu x0, 0 * 32(mem_ab); \
546	vmovdqu x1, 1 * 32(mem_ab); \
547	vmovdqu x2, 2 * 32(mem_ab); \
548	vmovdqu x3, 3 * 32(mem_ab); \
549	vmovdqu x4, 4 * 32(mem_ab); \
550	vmovdqu x5, 5 * 32(mem_ab); \
551	vmovdqu x6, 6 * 32(mem_ab); \
552	vmovdqu x7, 7 * 32(mem_ab); \
553	vmovdqu y0, 0 * 32(mem_cd); \
554	vmovdqu y1, 1 * 32(mem_cd); \
555	vmovdqu y2, 2 * 32(mem_cd); \
556	vmovdqu y3, 3 * 32(mem_cd); \
557	vmovdqu y4, 4 * 32(mem_cd); \
558	vmovdqu y5, 5 * 32(mem_cd); \
559	vmovdqu y6, 6 * 32(mem_cd); \
560	vmovdqu y7, 7 * 32(mem_cd);
561
562/* de-byteslice, apply post-whitening and store blocks */
563#define outunpack32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, \
564		    y5, y6, y7, key, stack_tmp0, stack_tmp1) \
565	byteslice_16x16b_fast(y0, y4, x0, x4, y1, y5, x1, x5, y2, y6, x2, x6, \
566			      y3, y7, x3, x7, stack_tmp0, stack_tmp1); \
567	\
568	vmovdqu x0, stack_tmp0; \
569	\
570	vpbroadcastq key, x0; \
571	vpshufb .Lpack_bswap, x0, x0; \
572	\
573	vpxor x0, y7, y7; \
574	vpxor x0, y6, y6; \
575	vpxor x0, y5, y5; \
576	vpxor x0, y4, y4; \
577	vpxor x0, y3, y3; \
578	vpxor x0, y2, y2; \
579	vpxor x0, y1, y1; \
580	vpxor x0, y0, y0; \
581	vpxor x0, x7, x7; \
582	vpxor x0, x6, x6; \
583	vpxor x0, x5, x5; \
584	vpxor x0, x4, x4; \
585	vpxor x0, x3, x3; \
586	vpxor x0, x2, x2; \
587	vpxor x0, x1, x1; \
588	vpxor stack_tmp0, x0, x0;
589
590#define write_output(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
591		     y6, y7, rio) \
592	vmovdqu x0, 0 * 32(rio); \
593	vmovdqu x1, 1 * 32(rio); \
594	vmovdqu x2, 2 * 32(rio); \
595	vmovdqu x3, 3 * 32(rio); \
596	vmovdqu x4, 4 * 32(rio); \
597	vmovdqu x5, 5 * 32(rio); \
598	vmovdqu x6, 6 * 32(rio); \
599	vmovdqu x7, 7 * 32(rio); \
600	vmovdqu y0, 8 * 32(rio); \
601	vmovdqu y1, 9 * 32(rio); \
602	vmovdqu y2, 10 * 32(rio); \
603	vmovdqu y3, 11 * 32(rio); \
604	vmovdqu y4, 12 * 32(rio); \
605	vmovdqu y5, 13 * 32(rio); \
606	vmovdqu y6, 14 * 32(rio); \
607	vmovdqu y7, 15 * 32(rio);
608
609
610.section	.rodata.cst32.shufb_16x16b, "aM", @progbits, 32
611.align 32
612#define SHUFB_BYTES(idx) \
613	0 + (idx), 4 + (idx), 8 + (idx), 12 + (idx)
614.Lshufb_16x16b:
615	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
616	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
617
618.section	.rodata.cst32.pack_bswap, "aM", @progbits, 32
619.align 32
620.Lpack_bswap:
621	.long 0x00010203, 0x04050607, 0x80808080, 0x80808080
622	.long 0x00010203, 0x04050607, 0x80808080, 0x80808080
623
624/* NB: section is mergeable, all elements must be aligned 16-byte blocks */
625.section	.rodata.cst16, "aM", @progbits, 16
626.align 16
627
628/* For CTR-mode IV byteswap */
629.Lbswap128_mask:
630	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
631
632/* For XTS mode */
633.Lxts_gf128mul_and_shl1_mask_0:
634	.byte 0x87, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0
635.Lxts_gf128mul_and_shl1_mask_1:
636	.byte 0x0e, 1, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0
637
638/*
639 * pre-SubByte transform
640 *
641 * pre-lookup for sbox1, sbox2, sbox3:
642 *   swap_bitendianness(
643 *       isom_map_camellia_to_aes(
644 *           camellia_f(
645 *               swap_bitendianess(in)
646 *           )
647 *       )
648 *   )
649 *
650 * (note: '⊕ 0xc5' inside camellia_f())
651 */
652.Lpre_tf_lo_s1:
653	.byte 0x45, 0xe8, 0x40, 0xed, 0x2e, 0x83, 0x2b, 0x86
654	.byte 0x4b, 0xe6, 0x4e, 0xe3, 0x20, 0x8d, 0x25, 0x88
655.Lpre_tf_hi_s1:
656	.byte 0x00, 0x51, 0xf1, 0xa0, 0x8a, 0xdb, 0x7b, 0x2a
657	.byte 0x09, 0x58, 0xf8, 0xa9, 0x83, 0xd2, 0x72, 0x23
658
659/*
660 * pre-SubByte transform
661 *
662 * pre-lookup for sbox4:
663 *   swap_bitendianness(
664 *       isom_map_camellia_to_aes(
665 *           camellia_f(
666 *               swap_bitendianess(in <<< 1)
667 *           )
668 *       )
669 *   )
670 *
671 * (note: '⊕ 0xc5' inside camellia_f())
672 */
673.Lpre_tf_lo_s4:
674	.byte 0x45, 0x40, 0x2e, 0x2b, 0x4b, 0x4e, 0x20, 0x25
675	.byte 0x14, 0x11, 0x7f, 0x7a, 0x1a, 0x1f, 0x71, 0x74
676.Lpre_tf_hi_s4:
677	.byte 0x00, 0xf1, 0x8a, 0x7b, 0x09, 0xf8, 0x83, 0x72
678	.byte 0xad, 0x5c, 0x27, 0xd6, 0xa4, 0x55, 0x2e, 0xdf
679
680/*
681 * post-SubByte transform
682 *
683 * post-lookup for sbox1, sbox4:
684 *  swap_bitendianness(
685 *      camellia_h(
686 *          isom_map_aes_to_camellia(
687 *              swap_bitendianness(
688 *                  aes_inverse_affine_transform(in)
689 *              )
690 *          )
691 *      )
692 *  )
693 *
694 * (note: '⊕ 0x6e' inside camellia_h())
695 */
696.Lpost_tf_lo_s1:
697	.byte 0x3c, 0xcc, 0xcf, 0x3f, 0x32, 0xc2, 0xc1, 0x31
698	.byte 0xdc, 0x2c, 0x2f, 0xdf, 0xd2, 0x22, 0x21, 0xd1
699.Lpost_tf_hi_s1:
700	.byte 0x00, 0xf9, 0x86, 0x7f, 0xd7, 0x2e, 0x51, 0xa8
701	.byte 0xa4, 0x5d, 0x22, 0xdb, 0x73, 0x8a, 0xf5, 0x0c
702
703/*
704 * post-SubByte transform
705 *
706 * post-lookup for sbox2:
707 *  swap_bitendianness(
708 *      camellia_h(
709 *          isom_map_aes_to_camellia(
710 *              swap_bitendianness(
711 *                  aes_inverse_affine_transform(in)
712 *              )
713 *          )
714 *      )
715 *  ) <<< 1
716 *
717 * (note: '⊕ 0x6e' inside camellia_h())
718 */
719.Lpost_tf_lo_s2:
720	.byte 0x78, 0x99, 0x9f, 0x7e, 0x64, 0x85, 0x83, 0x62
721	.byte 0xb9, 0x58, 0x5e, 0xbf, 0xa5, 0x44, 0x42, 0xa3
722.Lpost_tf_hi_s2:
723	.byte 0x00, 0xf3, 0x0d, 0xfe, 0xaf, 0x5c, 0xa2, 0x51
724	.byte 0x49, 0xba, 0x44, 0xb7, 0xe6, 0x15, 0xeb, 0x18
725
726/*
727 * post-SubByte transform
728 *
729 * post-lookup for sbox3:
730 *  swap_bitendianness(
731 *      camellia_h(
732 *          isom_map_aes_to_camellia(
733 *              swap_bitendianness(
734 *                  aes_inverse_affine_transform(in)
735 *              )
736 *          )
737 *      )
738 *  ) >>> 1
739 *
740 * (note: '⊕ 0x6e' inside camellia_h())
741 */
742.Lpost_tf_lo_s3:
743	.byte 0x1e, 0x66, 0xe7, 0x9f, 0x19, 0x61, 0xe0, 0x98
744	.byte 0x6e, 0x16, 0x97, 0xef, 0x69, 0x11, 0x90, 0xe8
745.Lpost_tf_hi_s3:
746	.byte 0x00, 0xfc, 0x43, 0xbf, 0xeb, 0x17, 0xa8, 0x54
747	.byte 0x52, 0xae, 0x11, 0xed, 0xb9, 0x45, 0xfa, 0x06
748
749/* For isolating SubBytes from AESENCLAST, inverse shift row */
750.Linv_shift_row:
751	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
752	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
753
754.section	.rodata.cst4.L0f0f0f0f, "aM", @progbits, 4
755.align 4
756/* 4-bit mask */
757.L0f0f0f0f:
758	.long 0x0f0f0f0f
759
760.text
761
762.align 8
763SYM_FUNC_START_LOCAL(__camellia_enc_blk32)
764	/* input:
765	 *	%rdi: ctx, CTX
766	 *	%rax: temporary storage, 512 bytes
767	 *	%ymm0..%ymm15: 32 plaintext blocks
768	 * output:
769	 *	%ymm0..%ymm15: 32 encrypted blocks, order swapped:
770	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
771	 */
772	FRAME_BEGIN
773
774	leaq 8 * 32(%rax), %rcx;
775
776	inpack32_post(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
777		      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
778		      %ymm15, %rax, %rcx);
779
780	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
781		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
782		     %ymm15, %rax, %rcx, 0);
783
784	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
785	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
786	      %ymm15,
787	      ((key_table + (8) * 8) + 0)(CTX),
788	      ((key_table + (8) * 8) + 4)(CTX),
789	      ((key_table + (8) * 8) + 8)(CTX),
790	      ((key_table + (8) * 8) + 12)(CTX));
791
792	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
793		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
794		     %ymm15, %rax, %rcx, 8);
795
796	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
797	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
798	      %ymm15,
799	      ((key_table + (16) * 8) + 0)(CTX),
800	      ((key_table + (16) * 8) + 4)(CTX),
801	      ((key_table + (16) * 8) + 8)(CTX),
802	      ((key_table + (16) * 8) + 12)(CTX));
803
804	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
805		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
806		     %ymm15, %rax, %rcx, 16);
807
808	movl $24, %r8d;
809	cmpl $16, key_length(CTX);
810	jne .Lenc_max32;
811
812.Lenc_done:
813	/* load CD for output */
814	vmovdqu 0 * 32(%rcx), %ymm8;
815	vmovdqu 1 * 32(%rcx), %ymm9;
816	vmovdqu 2 * 32(%rcx), %ymm10;
817	vmovdqu 3 * 32(%rcx), %ymm11;
818	vmovdqu 4 * 32(%rcx), %ymm12;
819	vmovdqu 5 * 32(%rcx), %ymm13;
820	vmovdqu 6 * 32(%rcx), %ymm14;
821	vmovdqu 7 * 32(%rcx), %ymm15;
822
823	outunpack32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
824		    %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
825		    %ymm15, (key_table)(CTX, %r8, 8), (%rax), 1 * 32(%rax));
826
827	FRAME_END
828	ret;
829
830.align 8
831.Lenc_max32:
832	movl $32, %r8d;
833
834	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
835	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
836	      %ymm15,
837	      ((key_table + (24) * 8) + 0)(CTX),
838	      ((key_table + (24) * 8) + 4)(CTX),
839	      ((key_table + (24) * 8) + 8)(CTX),
840	      ((key_table + (24) * 8) + 12)(CTX));
841
842	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
843		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
844		     %ymm15, %rax, %rcx, 24);
845
846	jmp .Lenc_done;
847SYM_FUNC_END(__camellia_enc_blk32)
848
849.align 8
850SYM_FUNC_START_LOCAL(__camellia_dec_blk32)
851	/* input:
852	 *	%rdi: ctx, CTX
853	 *	%rax: temporary storage, 512 bytes
854	 *	%r8d: 24 for 16 byte key, 32 for larger
855	 *	%ymm0..%ymm15: 16 encrypted blocks
856	 * output:
857	 *	%ymm0..%ymm15: 16 plaintext blocks, order swapped:
858	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
859	 */
860	FRAME_BEGIN
861
862	leaq 8 * 32(%rax), %rcx;
863
864	inpack32_post(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
865		      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
866		      %ymm15, %rax, %rcx);
867
868	cmpl $32, %r8d;
869	je .Ldec_max32;
870
871.Ldec_max24:
872	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
873		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
874		     %ymm15, %rax, %rcx, 16);
875
876	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
877	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
878	      %ymm15,
879	      ((key_table + (16) * 8) + 8)(CTX),
880	      ((key_table + (16) * 8) + 12)(CTX),
881	      ((key_table + (16) * 8) + 0)(CTX),
882	      ((key_table + (16) * 8) + 4)(CTX));
883
884	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
885		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
886		     %ymm15, %rax, %rcx, 8);
887
888	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
889	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
890	      %ymm15,
891	      ((key_table + (8) * 8) + 8)(CTX),
892	      ((key_table + (8) * 8) + 12)(CTX),
893	      ((key_table + (8) * 8) + 0)(CTX),
894	      ((key_table + (8) * 8) + 4)(CTX));
895
896	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
897		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
898		     %ymm15, %rax, %rcx, 0);
899
900	/* load CD for output */
901	vmovdqu 0 * 32(%rcx), %ymm8;
902	vmovdqu 1 * 32(%rcx), %ymm9;
903	vmovdqu 2 * 32(%rcx), %ymm10;
904	vmovdqu 3 * 32(%rcx), %ymm11;
905	vmovdqu 4 * 32(%rcx), %ymm12;
906	vmovdqu 5 * 32(%rcx), %ymm13;
907	vmovdqu 6 * 32(%rcx), %ymm14;
908	vmovdqu 7 * 32(%rcx), %ymm15;
909
910	outunpack32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
911		    %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
912		    %ymm15, (key_table)(CTX), (%rax), 1 * 32(%rax));
913
914	FRAME_END
915	ret;
916
917.align 8
918.Ldec_max32:
919	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
920		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
921		     %ymm15, %rax, %rcx, 24);
922
923	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
924	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
925	      %ymm15,
926	      ((key_table + (24) * 8) + 8)(CTX),
927	      ((key_table + (24) * 8) + 12)(CTX),
928	      ((key_table + (24) * 8) + 0)(CTX),
929	      ((key_table + (24) * 8) + 4)(CTX));
930
931	jmp .Ldec_max24;
932SYM_FUNC_END(__camellia_dec_blk32)
933
934SYM_FUNC_START(camellia_ecb_enc_32way)
935	/* input:
936	 *	%rdi: ctx, CTX
937	 *	%rsi: dst (32 blocks)
938	 *	%rdx: src (32 blocks)
939	 */
940	FRAME_BEGIN
941
942	vzeroupper;
943
944	inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
945		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
946		     %ymm15, %rdx, (key_table)(CTX));
947
948	/* now dst can be used as temporary buffer (even in src == dst case) */
949	movq	%rsi, %rax;
950
951	call __camellia_enc_blk32;
952
953	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
954		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
955		     %ymm8, %rsi);
956
957	vzeroupper;
958
959	FRAME_END
960	ret;
961SYM_FUNC_END(camellia_ecb_enc_32way)
962
963SYM_FUNC_START(camellia_ecb_dec_32way)
964	/* input:
965	 *	%rdi: ctx, CTX
966	 *	%rsi: dst (32 blocks)
967	 *	%rdx: src (32 blocks)
968	 */
969	FRAME_BEGIN
970
971	vzeroupper;
972
973	cmpl $16, key_length(CTX);
974	movl $32, %r8d;
975	movl $24, %eax;
976	cmovel %eax, %r8d; /* max */
977
978	inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
979		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
980		     %ymm15, %rdx, (key_table)(CTX, %r8, 8));
981
982	/* now dst can be used as temporary buffer (even in src == dst case) */
983	movq	%rsi, %rax;
984
985	call __camellia_dec_blk32;
986
987	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
988		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
989		     %ymm8, %rsi);
990
991	vzeroupper;
992
993	FRAME_END
994	ret;
995SYM_FUNC_END(camellia_ecb_dec_32way)
996
997SYM_FUNC_START(camellia_cbc_dec_32way)
998	/* input:
999	 *	%rdi: ctx, CTX
1000	 *	%rsi: dst (32 blocks)
1001	 *	%rdx: src (32 blocks)
1002	 */
1003	FRAME_BEGIN
1004
1005	vzeroupper;
1006
1007	cmpl $16, key_length(CTX);
1008	movl $32, %r8d;
1009	movl $24, %eax;
1010	cmovel %eax, %r8d; /* max */
1011
1012	inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
1013		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
1014		     %ymm15, %rdx, (key_table)(CTX, %r8, 8));
1015
1016	movq %rsp, %r10;
1017	cmpq %rsi, %rdx;
1018	je .Lcbc_dec_use_stack;
1019
1020	/* dst can be used as temporary storage, src is not overwritten. */
1021	movq %rsi, %rax;
1022	jmp .Lcbc_dec_continue;
1023
1024.Lcbc_dec_use_stack:
1025	/*
1026	 * dst still in-use (because dst == src), so use stack for temporary
1027	 * storage.
1028	 */
1029	subq $(16 * 32), %rsp;
1030	movq %rsp, %rax;
1031
1032.Lcbc_dec_continue:
1033	call __camellia_dec_blk32;
1034
1035	vmovdqu %ymm7, (%rax);
1036	vpxor %ymm7, %ymm7, %ymm7;
1037	vinserti128 $1, (%rdx), %ymm7, %ymm7;
1038	vpxor (%rax), %ymm7, %ymm7;
1039	movq %r10, %rsp;
1040	vpxor (0 * 32 + 16)(%rdx), %ymm6, %ymm6;
1041	vpxor (1 * 32 + 16)(%rdx), %ymm5, %ymm5;
1042	vpxor (2 * 32 + 16)(%rdx), %ymm4, %ymm4;
1043	vpxor (3 * 32 + 16)(%rdx), %ymm3, %ymm3;
1044	vpxor (4 * 32 + 16)(%rdx), %ymm2, %ymm2;
1045	vpxor (5 * 32 + 16)(%rdx), %ymm1, %ymm1;
1046	vpxor (6 * 32 + 16)(%rdx), %ymm0, %ymm0;
1047	vpxor (7 * 32 + 16)(%rdx), %ymm15, %ymm15;
1048	vpxor (8 * 32 + 16)(%rdx), %ymm14, %ymm14;
1049	vpxor (9 * 32 + 16)(%rdx), %ymm13, %ymm13;
1050	vpxor (10 * 32 + 16)(%rdx), %ymm12, %ymm12;
1051	vpxor (11 * 32 + 16)(%rdx), %ymm11, %ymm11;
1052	vpxor (12 * 32 + 16)(%rdx), %ymm10, %ymm10;
1053	vpxor (13 * 32 + 16)(%rdx), %ymm9, %ymm9;
1054	vpxor (14 * 32 + 16)(%rdx), %ymm8, %ymm8;
1055	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
1056		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
1057		     %ymm8, %rsi);
1058
1059	vzeroupper;
1060
1061	FRAME_END
1062	ret;
1063SYM_FUNC_END(camellia_cbc_dec_32way)
1064
1065#define inc_le128(x, minus_one, tmp) \
1066	vpcmpeqq minus_one, x, tmp; \
1067	vpsubq minus_one, x, x; \
1068	vpslldq $8, tmp, tmp; \
1069	vpsubq tmp, x, x;
1070
1071#define add2_le128(x, minus_one, minus_two, tmp1, tmp2) \
1072	vpcmpeqq minus_one, x, tmp1; \
1073	vpcmpeqq minus_two, x, tmp2; \
1074	vpsubq minus_two, x, x; \
1075	vpor tmp2, tmp1, tmp1; \
1076	vpslldq $8, tmp1, tmp1; \
1077	vpsubq tmp1, x, x;
1078
1079SYM_FUNC_START(camellia_ctr_32way)
1080	/* input:
1081	 *	%rdi: ctx, CTX
1082	 *	%rsi: dst (32 blocks)
1083	 *	%rdx: src (32 blocks)
1084	 *	%rcx: iv (little endian, 128bit)
1085	 */
1086	FRAME_BEGIN
1087
1088	vzeroupper;
1089
1090	movq %rsp, %r10;
1091	cmpq %rsi, %rdx;
1092	je .Lctr_use_stack;
1093
1094	/* dst can be used as temporary storage, src is not overwritten. */
1095	movq %rsi, %rax;
1096	jmp .Lctr_continue;
1097
1098.Lctr_use_stack:
1099	subq $(16 * 32), %rsp;
1100	movq %rsp, %rax;
1101
1102.Lctr_continue:
1103	vpcmpeqd %ymm15, %ymm15, %ymm15;
1104	vpsrldq $8, %ymm15, %ymm15; /* ab: -1:0 ; cd: -1:0 */
1105	vpaddq %ymm15, %ymm15, %ymm12; /* ab: -2:0 ; cd: -2:0 */
1106
1107	/* load IV and byteswap */
1108	vmovdqu (%rcx), %xmm0;
1109	vmovdqa %xmm0, %xmm1;
1110	inc_le128(%xmm0, %xmm15, %xmm14);
1111	vbroadcasti128 .Lbswap128_mask, %ymm14;
1112	vinserti128 $1, %xmm0, %ymm1, %ymm0;
1113	vpshufb %ymm14, %ymm0, %ymm13;
1114	vmovdqu %ymm13, 15 * 32(%rax);
1115
1116	/* construct IVs */
1117	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13); /* ab:le2 ; cd:le3 */
1118	vpshufb %ymm14, %ymm0, %ymm13;
1119	vmovdqu %ymm13, 14 * 32(%rax);
1120	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1121	vpshufb %ymm14, %ymm0, %ymm13;
1122	vmovdqu %ymm13, 13 * 32(%rax);
1123	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1124	vpshufb %ymm14, %ymm0, %ymm13;
1125	vmovdqu %ymm13, 12 * 32(%rax);
1126	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1127	vpshufb %ymm14, %ymm0, %ymm13;
1128	vmovdqu %ymm13, 11 * 32(%rax);
1129	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1130	vpshufb %ymm14, %ymm0, %ymm10;
1131	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1132	vpshufb %ymm14, %ymm0, %ymm9;
1133	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1134	vpshufb %ymm14, %ymm0, %ymm8;
1135	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1136	vpshufb %ymm14, %ymm0, %ymm7;
1137	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1138	vpshufb %ymm14, %ymm0, %ymm6;
1139	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1140	vpshufb %ymm14, %ymm0, %ymm5;
1141	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1142	vpshufb %ymm14, %ymm0, %ymm4;
1143	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1144	vpshufb %ymm14, %ymm0, %ymm3;
1145	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1146	vpshufb %ymm14, %ymm0, %ymm2;
1147	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1148	vpshufb %ymm14, %ymm0, %ymm1;
1149	add2_le128(%ymm0, %ymm15, %ymm12, %ymm11, %ymm13);
1150	vextracti128 $1, %ymm0, %xmm13;
1151	vpshufb %ymm14, %ymm0, %ymm0;
1152	inc_le128(%xmm13, %xmm15, %xmm14);
1153	vmovdqu %xmm13, (%rcx);
1154
1155	/* inpack32_pre: */
1156	vpbroadcastq (key_table)(CTX), %ymm15;
1157	vpshufb .Lpack_bswap, %ymm15, %ymm15;
1158	vpxor %ymm0, %ymm15, %ymm0;
1159	vpxor %ymm1, %ymm15, %ymm1;
1160	vpxor %ymm2, %ymm15, %ymm2;
1161	vpxor %ymm3, %ymm15, %ymm3;
1162	vpxor %ymm4, %ymm15, %ymm4;
1163	vpxor %ymm5, %ymm15, %ymm5;
1164	vpxor %ymm6, %ymm15, %ymm6;
1165	vpxor %ymm7, %ymm15, %ymm7;
1166	vpxor %ymm8, %ymm15, %ymm8;
1167	vpxor %ymm9, %ymm15, %ymm9;
1168	vpxor %ymm10, %ymm15, %ymm10;
1169	vpxor 11 * 32(%rax), %ymm15, %ymm11;
1170	vpxor 12 * 32(%rax), %ymm15, %ymm12;
1171	vpxor 13 * 32(%rax), %ymm15, %ymm13;
1172	vpxor 14 * 32(%rax), %ymm15, %ymm14;
1173	vpxor 15 * 32(%rax), %ymm15, %ymm15;
1174
1175	call __camellia_enc_blk32;
1176
1177	movq %r10, %rsp;
1178
1179	vpxor 0 * 32(%rdx), %ymm7, %ymm7;
1180	vpxor 1 * 32(%rdx), %ymm6, %ymm6;
1181	vpxor 2 * 32(%rdx), %ymm5, %ymm5;
1182	vpxor 3 * 32(%rdx), %ymm4, %ymm4;
1183	vpxor 4 * 32(%rdx), %ymm3, %ymm3;
1184	vpxor 5 * 32(%rdx), %ymm2, %ymm2;
1185	vpxor 6 * 32(%rdx), %ymm1, %ymm1;
1186	vpxor 7 * 32(%rdx), %ymm0, %ymm0;
1187	vpxor 8 * 32(%rdx), %ymm15, %ymm15;
1188	vpxor 9 * 32(%rdx), %ymm14, %ymm14;
1189	vpxor 10 * 32(%rdx), %ymm13, %ymm13;
1190	vpxor 11 * 32(%rdx), %ymm12, %ymm12;
1191	vpxor 12 * 32(%rdx), %ymm11, %ymm11;
1192	vpxor 13 * 32(%rdx), %ymm10, %ymm10;
1193	vpxor 14 * 32(%rdx), %ymm9, %ymm9;
1194	vpxor 15 * 32(%rdx), %ymm8, %ymm8;
1195	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
1196		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
1197		     %ymm8, %rsi);
1198
1199	vzeroupper;
1200
1201	FRAME_END
1202	ret;
1203SYM_FUNC_END(camellia_ctr_32way)
1204
1205#define gf128mul_x_ble(iv, mask, tmp) \
1206	vpsrad $31, iv, tmp; \
1207	vpaddq iv, iv, iv; \
1208	vpshufd $0x13, tmp, tmp; \
1209	vpand mask, tmp, tmp; \
1210	vpxor tmp, iv, iv;
1211
1212#define gf128mul_x2_ble(iv, mask1, mask2, tmp0, tmp1) \
1213	vpsrad $31, iv, tmp0; \
1214	vpaddq iv, iv, tmp1; \
1215	vpsllq $2, iv, iv; \
1216	vpshufd $0x13, tmp0, tmp0; \
1217	vpsrad $31, tmp1, tmp1; \
1218	vpand mask2, tmp0, tmp0; \
1219	vpshufd $0x13, tmp1, tmp1; \
1220	vpxor tmp0, iv, iv; \
1221	vpand mask1, tmp1, tmp1; \
1222	vpxor tmp1, iv, iv;
1223
1224.align 8
1225SYM_FUNC_START_LOCAL(camellia_xts_crypt_32way)
1226	/* input:
1227	 *	%rdi: ctx, CTX
1228	 *	%rsi: dst (32 blocks)
1229	 *	%rdx: src (32 blocks)
1230	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
1231	 *	%r8: index for input whitening key
1232	 *	%r9: pointer to  __camellia_enc_blk32 or __camellia_dec_blk32
1233	 */
1234	FRAME_BEGIN
1235
1236	vzeroupper;
1237
1238	subq $(16 * 32), %rsp;
1239	movq %rsp, %rax;
1240
1241	vbroadcasti128 .Lxts_gf128mul_and_shl1_mask_0, %ymm12;
1242
1243	/* load IV and construct second IV */
1244	vmovdqu (%rcx), %xmm0;
1245	vmovdqa %xmm0, %xmm15;
1246	gf128mul_x_ble(%xmm0, %xmm12, %xmm13);
1247	vbroadcasti128 .Lxts_gf128mul_and_shl1_mask_1, %ymm13;
1248	vinserti128 $1, %xmm0, %ymm15, %ymm0;
1249	vpxor 0 * 32(%rdx), %ymm0, %ymm15;
1250	vmovdqu %ymm15, 15 * 32(%rax);
1251	vmovdqu %ymm0, 0 * 32(%rsi);
1252
1253	/* construct IVs */
1254	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1255	vpxor 1 * 32(%rdx), %ymm0, %ymm15;
1256	vmovdqu %ymm15, 14 * 32(%rax);
1257	vmovdqu %ymm0, 1 * 32(%rsi);
1258
1259	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1260	vpxor 2 * 32(%rdx), %ymm0, %ymm15;
1261	vmovdqu %ymm15, 13 * 32(%rax);
1262	vmovdqu %ymm0, 2 * 32(%rsi);
1263
1264	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1265	vpxor 3 * 32(%rdx), %ymm0, %ymm15;
1266	vmovdqu %ymm15, 12 * 32(%rax);
1267	vmovdqu %ymm0, 3 * 32(%rsi);
1268
1269	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1270	vpxor 4 * 32(%rdx), %ymm0, %ymm11;
1271	vmovdqu %ymm0, 4 * 32(%rsi);
1272
1273	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1274	vpxor 5 * 32(%rdx), %ymm0, %ymm10;
1275	vmovdqu %ymm0, 5 * 32(%rsi);
1276
1277	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1278	vpxor 6 * 32(%rdx), %ymm0, %ymm9;
1279	vmovdqu %ymm0, 6 * 32(%rsi);
1280
1281	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1282	vpxor 7 * 32(%rdx), %ymm0, %ymm8;
1283	vmovdqu %ymm0, 7 * 32(%rsi);
1284
1285	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1286	vpxor 8 * 32(%rdx), %ymm0, %ymm7;
1287	vmovdqu %ymm0, 8 * 32(%rsi);
1288
1289	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1290	vpxor 9 * 32(%rdx), %ymm0, %ymm6;
1291	vmovdqu %ymm0, 9 * 32(%rsi);
1292
1293	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1294	vpxor 10 * 32(%rdx), %ymm0, %ymm5;
1295	vmovdqu %ymm0, 10 * 32(%rsi);
1296
1297	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1298	vpxor 11 * 32(%rdx), %ymm0, %ymm4;
1299	vmovdqu %ymm0, 11 * 32(%rsi);
1300
1301	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1302	vpxor 12 * 32(%rdx), %ymm0, %ymm3;
1303	vmovdqu %ymm0, 12 * 32(%rsi);
1304
1305	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1306	vpxor 13 * 32(%rdx), %ymm0, %ymm2;
1307	vmovdqu %ymm0, 13 * 32(%rsi);
1308
1309	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1310	vpxor 14 * 32(%rdx), %ymm0, %ymm1;
1311	vmovdqu %ymm0, 14 * 32(%rsi);
1312
1313	gf128mul_x2_ble(%ymm0, %ymm12, %ymm13, %ymm14, %ymm15);
1314	vpxor 15 * 32(%rdx), %ymm0, %ymm15;
1315	vmovdqu %ymm15, 0 * 32(%rax);
1316	vmovdqu %ymm0, 15 * 32(%rsi);
1317
1318	vextracti128 $1, %ymm0, %xmm0;
1319	gf128mul_x_ble(%xmm0, %xmm12, %xmm15);
1320	vmovdqu %xmm0, (%rcx);
1321
1322	/* inpack32_pre: */
1323	vpbroadcastq (key_table)(CTX, %r8, 8), %ymm15;
1324	vpshufb .Lpack_bswap, %ymm15, %ymm15;
1325	vpxor 0 * 32(%rax), %ymm15, %ymm0;
1326	vpxor %ymm1, %ymm15, %ymm1;
1327	vpxor %ymm2, %ymm15, %ymm2;
1328	vpxor %ymm3, %ymm15, %ymm3;
1329	vpxor %ymm4, %ymm15, %ymm4;
1330	vpxor %ymm5, %ymm15, %ymm5;
1331	vpxor %ymm6, %ymm15, %ymm6;
1332	vpxor %ymm7, %ymm15, %ymm7;
1333	vpxor %ymm8, %ymm15, %ymm8;
1334	vpxor %ymm9, %ymm15, %ymm9;
1335	vpxor %ymm10, %ymm15, %ymm10;
1336	vpxor %ymm11, %ymm15, %ymm11;
1337	vpxor 12 * 32(%rax), %ymm15, %ymm12;
1338	vpxor 13 * 32(%rax), %ymm15, %ymm13;
1339	vpxor 14 * 32(%rax), %ymm15, %ymm14;
1340	vpxor 15 * 32(%rax), %ymm15, %ymm15;
1341
1342	CALL_NOSPEC %r9;
1343
1344	addq $(16 * 32), %rsp;
1345
1346	vpxor 0 * 32(%rsi), %ymm7, %ymm7;
1347	vpxor 1 * 32(%rsi), %ymm6, %ymm6;
1348	vpxor 2 * 32(%rsi), %ymm5, %ymm5;
1349	vpxor 3 * 32(%rsi), %ymm4, %ymm4;
1350	vpxor 4 * 32(%rsi), %ymm3, %ymm3;
1351	vpxor 5 * 32(%rsi), %ymm2, %ymm2;
1352	vpxor 6 * 32(%rsi), %ymm1, %ymm1;
1353	vpxor 7 * 32(%rsi), %ymm0, %ymm0;
1354	vpxor 8 * 32(%rsi), %ymm15, %ymm15;
1355	vpxor 9 * 32(%rsi), %ymm14, %ymm14;
1356	vpxor 10 * 32(%rsi), %ymm13, %ymm13;
1357	vpxor 11 * 32(%rsi), %ymm12, %ymm12;
1358	vpxor 12 * 32(%rsi), %ymm11, %ymm11;
1359	vpxor 13 * 32(%rsi), %ymm10, %ymm10;
1360	vpxor 14 * 32(%rsi), %ymm9, %ymm9;
1361	vpxor 15 * 32(%rsi), %ymm8, %ymm8;
1362	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
1363		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
1364		     %ymm8, %rsi);
1365
1366	vzeroupper;
1367
1368	FRAME_END
1369	ret;
1370SYM_FUNC_END(camellia_xts_crypt_32way)
1371
1372SYM_FUNC_START(camellia_xts_enc_32way)
1373	/* input:
1374	 *	%rdi: ctx, CTX
1375	 *	%rsi: dst (32 blocks)
1376	 *	%rdx: src (32 blocks)
1377	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
1378	 */
1379
1380	xorl %r8d, %r8d; /* input whitening key, 0 for enc */
1381
1382	leaq __camellia_enc_blk32, %r9;
1383
1384	jmp camellia_xts_crypt_32way;
1385SYM_FUNC_END(camellia_xts_enc_32way)
1386
1387SYM_FUNC_START(camellia_xts_dec_32way)
1388	/* input:
1389	 *	%rdi: ctx, CTX
1390	 *	%rsi: dst (32 blocks)
1391	 *	%rdx: src (32 blocks)
1392	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
1393	 */
1394
1395	cmpl $16, key_length(CTX);
1396	movl $32, %r8d;
1397	movl $24, %eax;
1398	cmovel %eax, %r8d;  /* input whitening key, last for dec */
1399
1400	leaq __camellia_dec_blk32, %r9;
1401
1402	jmp camellia_xts_crypt_32way;
1403SYM_FUNC_END(camellia_xts_dec_32way)
1404