1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * Twofish Cipher 8-way parallel algorithm (AVX/x86_64)
4 *
5 * Copyright (C) 2012 Johannes Goetzfried
6 *     <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
7 *
8 * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
9 */
10
11#include <linux/linkage.h>
12#include <asm/frame.h>
13#include "glue_helper-asm-avx.S"
14
15.file "twofish-avx-x86_64-asm_64.S"
16
17.section	.rodata.cst16.bswap128_mask, "aM", @progbits, 16
18.align 16
19.Lbswap128_mask:
20	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
21
22.text
23
24/* structure of crypto context */
25#define s0	0
26#define s1	1024
27#define s2	2048
28#define s3	3072
29#define w	4096
30#define k	4128
31
32/**********************************************************************
33  8-way AVX twofish
34 **********************************************************************/
35#define CTX %rdi
36
37#define RA1 %xmm0
38#define RB1 %xmm1
39#define RC1 %xmm2
40#define RD1 %xmm3
41
42#define RA2 %xmm4
43#define RB2 %xmm5
44#define RC2 %xmm6
45#define RD2 %xmm7
46
47#define RX0 %xmm8
48#define RY0 %xmm9
49
50#define RX1 %xmm10
51#define RY1 %xmm11
52
53#define RK1 %xmm12
54#define RK2 %xmm13
55
56#define RT %xmm14
57#define RR %xmm15
58
59#define RID1  %r13
60#define RID1d %r13d
61#define RID2  %rsi
62#define RID2d %esi
63
64#define RGI1   %rdx
65#define RGI1bl %dl
66#define RGI1bh %dh
67#define RGI2   %rcx
68#define RGI2bl %cl
69#define RGI2bh %ch
70
71#define RGI3   %rax
72#define RGI3bl %al
73#define RGI3bh %ah
74#define RGI4   %rbx
75#define RGI4bl %bl
76#define RGI4bh %bh
77
78#define RGS1  %r8
79#define RGS1d %r8d
80#define RGS2  %r9
81#define RGS2d %r9d
82#define RGS3  %r10
83#define RGS3d %r10d
84
85
86#define lookup_32bit(t0, t1, t2, t3, src, dst, interleave_op, il_reg) \
87	movzbl		src ## bl,        RID1d;     \
88	movzbl		src ## bh,        RID2d;     \
89	shrq $16,	src;                         \
90	movl		t0(CTX, RID1, 4), dst ## d;  \
91	movl		t1(CTX, RID2, 4), RID2d;     \
92	movzbl		src ## bl,        RID1d;     \
93	xorl		RID2d,            dst ## d;  \
94	movzbl		src ## bh,        RID2d;     \
95	interleave_op(il_reg);			     \
96	xorl		t2(CTX, RID1, 4), dst ## d;  \
97	xorl		t3(CTX, RID2, 4), dst ## d;
98
99#define dummy(d) /* do nothing */
100
101#define shr_next(reg) \
102	shrq $16,	reg;
103
104#define G(gi1, gi2, x, t0, t1, t2, t3) \
105	lookup_32bit(t0, t1, t2, t3, ##gi1, RGS1, shr_next, ##gi1);  \
106	lookup_32bit(t0, t1, t2, t3, ##gi2, RGS3, shr_next, ##gi2);  \
107	\
108	lookup_32bit(t0, t1, t2, t3, ##gi1, RGS2, dummy, none);      \
109	shlq $32,	RGS2;                                        \
110	orq		RGS1, RGS2;                                  \
111	lookup_32bit(t0, t1, t2, t3, ##gi2, RGS1, dummy, none);      \
112	shlq $32,	RGS1;                                        \
113	orq		RGS1, RGS3;
114
115#define round_head_2(a, b, x1, y1, x2, y2) \
116	vmovq		b ## 1, RGI3;           \
117	vpextrq $1,	b ## 1, RGI4;           \
118	\
119	G(RGI1, RGI2, x1, s0, s1, s2, s3);      \
120	vmovq		a ## 2, RGI1;           \
121	vpextrq $1,	a ## 2, RGI2;           \
122	vmovq		RGS2, x1;               \
123	vpinsrq $1,	RGS3, x1, x1;           \
124	\
125	G(RGI3, RGI4, y1, s1, s2, s3, s0);      \
126	vmovq		b ## 2, RGI3;           \
127	vpextrq $1,	b ## 2, RGI4;           \
128	vmovq		RGS2, y1;               \
129	vpinsrq $1,	RGS3, y1, y1;           \
130	\
131	G(RGI1, RGI2, x2, s0, s1, s2, s3);      \
132	vmovq		RGS2, x2;               \
133	vpinsrq $1,	RGS3, x2, x2;           \
134	\
135	G(RGI3, RGI4, y2, s1, s2, s3, s0);      \
136	vmovq		RGS2, y2;               \
137	vpinsrq $1,	RGS3, y2, y2;
138
139#define encround_tail(a, b, c, d, x, y, prerotate) \
140	vpaddd			x, y,   x; \
141	vpaddd			x, RK1, RT;\
142	prerotate(b);			   \
143	vpxor			RT, c,  c; \
144	vpaddd			y, x,   y; \
145	vpaddd			y, RK2, y; \
146	vpsrld $1,		c, RT;     \
147	vpslld $(32 - 1),	c, c;      \
148	vpor			c, RT,  c; \
149	vpxor			d, y,   d; \
150
151#define decround_tail(a, b, c, d, x, y, prerotate) \
152	vpaddd			x, y,   x; \
153	vpaddd			x, RK1, RT;\
154	prerotate(a);			   \
155	vpxor			RT, c,  c; \
156	vpaddd			y, x,   y; \
157	vpaddd			y, RK2, y; \
158	vpxor			d, y,   d; \
159	vpsrld $1,		d, y;      \
160	vpslld $(32 - 1),	d, d;      \
161	vpor			d, y,   d; \
162
163#define rotate_1l(x) \
164	vpslld $1,		x, RR;     \
165	vpsrld $(32 - 1),	x, x;      \
166	vpor			x, RR,  x;
167
168#define preload_rgi(c) \
169	vmovq			c, RGI1; \
170	vpextrq $1,		c, RGI2;
171
172#define encrypt_round(n, a, b, c, d, preload, prerotate) \
173	vbroadcastss (k+4*(2*(n)))(CTX),   RK1;                  \
174	vbroadcastss (k+4*(2*(n)+1))(CTX), RK2;                  \
175	round_head_2(a, b, RX0, RY0, RX1, RY1);                  \
176	encround_tail(a ## 1, b ## 1, c ## 1, d ## 1, RX0, RY0, prerotate); \
177	preload(c ## 1);                                         \
178	encround_tail(a ## 2, b ## 2, c ## 2, d ## 2, RX1, RY1, prerotate);
179
180#define decrypt_round(n, a, b, c, d, preload, prerotate) \
181	vbroadcastss (k+4*(2*(n)))(CTX),   RK1;                  \
182	vbroadcastss (k+4*(2*(n)+1))(CTX), RK2;                  \
183	round_head_2(a, b, RX0, RY0, RX1, RY1);                  \
184	decround_tail(a ## 1, b ## 1, c ## 1, d ## 1, RX0, RY0, prerotate); \
185	preload(c ## 1);                                         \
186	decround_tail(a ## 2, b ## 2, c ## 2, d ## 2, RX1, RY1, prerotate);
187
188#define encrypt_cycle(n) \
189	encrypt_round((2*n), RA, RB, RC, RD, preload_rgi, rotate_1l); \
190	encrypt_round(((2*n) + 1), RC, RD, RA, RB, preload_rgi, rotate_1l);
191
192#define encrypt_cycle_last(n) \
193	encrypt_round((2*n), RA, RB, RC, RD, preload_rgi, rotate_1l); \
194	encrypt_round(((2*n) + 1), RC, RD, RA, RB, dummy, dummy);
195
196#define decrypt_cycle(n) \
197	decrypt_round(((2*n) + 1), RC, RD, RA, RB, preload_rgi, rotate_1l); \
198	decrypt_round((2*n), RA, RB, RC, RD, preload_rgi, rotate_1l);
199
200#define decrypt_cycle_last(n) \
201	decrypt_round(((2*n) + 1), RC, RD, RA, RB, preload_rgi, rotate_1l); \
202	decrypt_round((2*n), RA, RB, RC, RD, dummy, dummy);
203
204#define transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
205	vpunpckldq		x1, x0, t0; \
206	vpunpckhdq		x1, x0, t2; \
207	vpunpckldq		x3, x2, t1; \
208	vpunpckhdq		x3, x2, x3; \
209	\
210	vpunpcklqdq		t1, t0, x0; \
211	vpunpckhqdq		t1, t0, x1; \
212	vpunpcklqdq		x3, t2, x2; \
213	vpunpckhqdq		x3, t2, x3;
214
215#define inpack_blocks(x0, x1, x2, x3, wkey, t0, t1, t2) \
216	vpxor		x0, wkey, x0; \
217	vpxor		x1, wkey, x1; \
218	vpxor		x2, wkey, x2; \
219	vpxor		x3, wkey, x3; \
220	\
221	transpose_4x4(x0, x1, x2, x3, t0, t1, t2)
222
223#define outunpack_blocks(x0, x1, x2, x3, wkey, t0, t1, t2) \
224	transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
225	\
226	vpxor		x0, wkey, x0; \
227	vpxor		x1, wkey, x1; \
228	vpxor		x2, wkey, x2; \
229	vpxor		x3, wkey, x3;
230
231.align 8
232SYM_FUNC_START_LOCAL(__twofish_enc_blk8)
233	/* input:
234	 *	%rdi: ctx, CTX
235	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: blocks
236	 * output:
237	 *	RC1, RD1, RA1, RB1, RC2, RD2, RA2, RB2: encrypted blocks
238	 */
239
240	vmovdqu w(CTX), RK1;
241
242	pushq %r13;
243	pushq %rbx;
244	pushq %rcx;
245
246	inpack_blocks(RA1, RB1, RC1, RD1, RK1, RX0, RY0, RK2);
247	preload_rgi(RA1);
248	rotate_1l(RD1);
249	inpack_blocks(RA2, RB2, RC2, RD2, RK1, RX0, RY0, RK2);
250	rotate_1l(RD2);
251
252	encrypt_cycle(0);
253	encrypt_cycle(1);
254	encrypt_cycle(2);
255	encrypt_cycle(3);
256	encrypt_cycle(4);
257	encrypt_cycle(5);
258	encrypt_cycle(6);
259	encrypt_cycle_last(7);
260
261	vmovdqu (w+4*4)(CTX), RK1;
262
263	popq %rcx;
264	popq %rbx;
265	popq %r13;
266
267	outunpack_blocks(RC1, RD1, RA1, RB1, RK1, RX0, RY0, RK2);
268	outunpack_blocks(RC2, RD2, RA2, RB2, RK1, RX0, RY0, RK2);
269
270	RET;
271SYM_FUNC_END(__twofish_enc_blk8)
272
273.align 8
274SYM_FUNC_START_LOCAL(__twofish_dec_blk8)
275	/* input:
276	 *	%rdi: ctx, CTX
277	 *	RC1, RD1, RA1, RB1, RC2, RD2, RA2, RB2: encrypted blocks
278	 * output:
279	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: decrypted blocks
280	 */
281
282	vmovdqu (w+4*4)(CTX), RK1;
283
284	pushq %r13;
285	pushq %rbx;
286
287	inpack_blocks(RC1, RD1, RA1, RB1, RK1, RX0, RY0, RK2);
288	preload_rgi(RC1);
289	rotate_1l(RA1);
290	inpack_blocks(RC2, RD2, RA2, RB2, RK1, RX0, RY0, RK2);
291	rotate_1l(RA2);
292
293	decrypt_cycle(7);
294	decrypt_cycle(6);
295	decrypt_cycle(5);
296	decrypt_cycle(4);
297	decrypt_cycle(3);
298	decrypt_cycle(2);
299	decrypt_cycle(1);
300	decrypt_cycle_last(0);
301
302	vmovdqu (w)(CTX), RK1;
303
304	popq %rbx;
305	popq %r13;
306
307	outunpack_blocks(RA1, RB1, RC1, RD1, RK1, RX0, RY0, RK2);
308	outunpack_blocks(RA2, RB2, RC2, RD2, RK1, RX0, RY0, RK2);
309
310	RET;
311SYM_FUNC_END(__twofish_dec_blk8)
312
313SYM_FUNC_START(twofish_ecb_enc_8way)
314	/* input:
315	 *	%rdi: ctx, CTX
316	 *	%rsi: dst
317	 *	%rdx: src
318	 */
319	FRAME_BEGIN
320
321	movq %rsi, %r11;
322
323	load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
324
325	call __twofish_enc_blk8;
326
327	store_8way(%r11, RC1, RD1, RA1, RB1, RC2, RD2, RA2, RB2);
328
329	FRAME_END
330	RET;
331SYM_FUNC_END(twofish_ecb_enc_8way)
332
333SYM_FUNC_START(twofish_ecb_dec_8way)
334	/* input:
335	 *	%rdi: ctx, CTX
336	 *	%rsi: dst
337	 *	%rdx: src
338	 */
339	FRAME_BEGIN
340
341	movq %rsi, %r11;
342
343	load_8way(%rdx, RC1, RD1, RA1, RB1, RC2, RD2, RA2, RB2);
344
345	call __twofish_dec_blk8;
346
347	store_8way(%r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
348
349	FRAME_END
350	RET;
351SYM_FUNC_END(twofish_ecb_dec_8way)
352
353SYM_FUNC_START(twofish_cbc_dec_8way)
354	/* input:
355	 *	%rdi: ctx, CTX
356	 *	%rsi: dst
357	 *	%rdx: src
358	 */
359	FRAME_BEGIN
360
361	pushq %r12;
362
363	movq %rsi, %r11;
364	movq %rdx, %r12;
365
366	load_8way(%rdx, RC1, RD1, RA1, RB1, RC2, RD2, RA2, RB2);
367
368	call __twofish_dec_blk8;
369
370	store_cbc_8way(%r12, %r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
371
372	popq %r12;
373
374	FRAME_END
375	RET;
376SYM_FUNC_END(twofish_cbc_dec_8way)
377