1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * Blowfish Cipher Algorithm (x86_64)
4 *
5 * Copyright (C) 2011 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
6 */
7
8#include <linux/linkage.h>
9
10.file "blowfish-x86_64-asm.S"
11.text
12
13/* structure of crypto context */
14#define p	0
15#define s0	((16 + 2) * 4)
16#define s1	((16 + 2 + (1 * 256)) * 4)
17#define s2	((16 + 2 + (2 * 256)) * 4)
18#define s3	((16 + 2 + (3 * 256)) * 4)
19
20/* register macros */
21#define CTX %r12
22#define RIO %rsi
23
24#define RX0 %rax
25#define RX1 %rbx
26#define RX2 %rcx
27#define RX3 %rdx
28
29#define RX0d %eax
30#define RX1d %ebx
31#define RX2d %ecx
32#define RX3d %edx
33
34#define RX0bl %al
35#define RX1bl %bl
36#define RX2bl %cl
37#define RX3bl %dl
38
39#define RX0bh %ah
40#define RX1bh %bh
41#define RX2bh %ch
42#define RX3bh %dh
43
44#define RT0 %rdi
45#define RT1 %rsi
46#define RT2 %r8
47#define RT3 %r9
48
49#define RT0d %edi
50#define RT1d %esi
51#define RT2d %r8d
52#define RT3d %r9d
53
54#define RKEY %r10
55
56/***********************************************************************
57 * 1-way blowfish
58 ***********************************************************************/
59#define F() \
60	rorq $16,		RX0; \
61	movzbl RX0bh,		RT0d; \
62	movzbl RX0bl,		RT1d; \
63	rolq $16,		RX0; \
64	movl s0(CTX,RT0,4),	RT0d; \
65	addl s1(CTX,RT1,4),	RT0d; \
66	movzbl RX0bh,		RT1d; \
67	movzbl RX0bl,		RT2d; \
68	rolq $32,		RX0; \
69	xorl s2(CTX,RT1,4),	RT0d; \
70	addl s3(CTX,RT2,4),	RT0d; \
71	xorq RT0,		RX0;
72
73#define add_roundkey_enc(n) \
74	xorq p+4*(n)(CTX), 	RX0;
75
76#define round_enc(n) \
77	add_roundkey_enc(n); \
78	\
79	F(); \
80	F();
81
82#define add_roundkey_dec(n) \
83	movq p+4*(n-1)(CTX),	RT0; \
84	rorq $32,		RT0; \
85	xorq RT0,		RX0;
86
87#define round_dec(n) \
88	add_roundkey_dec(n); \
89	\
90	F(); \
91	F(); \
92
93#define read_block() \
94	movq (RIO), 		RX0; \
95	rorq $32, 		RX0; \
96	bswapq 			RX0;
97
98#define write_block() \
99	bswapq 			RX0; \
100	movq RX0, 		(RIO);
101
102#define xor_block() \
103	bswapq 			RX0; \
104	xorq RX0, 		(RIO);
105
106ENTRY(__blowfish_enc_blk)
107	/* input:
108	 *	%rdi: ctx
109	 *	%rsi: dst
110	 *	%rdx: src
111	 *	%rcx: bool, if true: xor output
112	 */
113	movq %r12, %r11;
114
115	movq %rdi, CTX;
116	movq %rsi, %r10;
117	movq %rdx, RIO;
118
119	read_block();
120
121	round_enc(0);
122	round_enc(2);
123	round_enc(4);
124	round_enc(6);
125	round_enc(8);
126	round_enc(10);
127	round_enc(12);
128	round_enc(14);
129	add_roundkey_enc(16);
130
131	movq %r11, %r12;
132
133	movq %r10, RIO;
134	test %cl, %cl;
135	jnz .L__enc_xor;
136
137	write_block();
138	ret;
139.L__enc_xor:
140	xor_block();
141	ret;
142ENDPROC(__blowfish_enc_blk)
143
144ENTRY(blowfish_dec_blk)
145	/* input:
146	 *	%rdi: ctx
147	 *	%rsi: dst
148	 *	%rdx: src
149	 */
150	movq %r12, %r11;
151
152	movq %rdi, CTX;
153	movq %rsi, %r10;
154	movq %rdx, RIO;
155
156	read_block();
157
158	round_dec(17);
159	round_dec(15);
160	round_dec(13);
161	round_dec(11);
162	round_dec(9);
163	round_dec(7);
164	round_dec(5);
165	round_dec(3);
166	add_roundkey_dec(1);
167
168	movq %r10, RIO;
169	write_block();
170
171	movq %r11, %r12;
172
173	ret;
174ENDPROC(blowfish_dec_blk)
175
176/**********************************************************************
177  4-way blowfish, four blocks parallel
178 **********************************************************************/
179
180/* F() for 4-way. Slower when used alone/1-way, but faster when used
181 * parallel/4-way (tested on AMD Phenom II & Intel Xeon E7330).
182 */
183#define F4(x) \
184	movzbl x ## bh,		RT1d; \
185	movzbl x ## bl,		RT3d; \
186	rorq $16,		x; \
187	movzbl x ## bh,		RT0d; \
188	movzbl x ## bl,		RT2d; \
189	rorq $16,		x; \
190	movl s0(CTX,RT0,4),	RT0d; \
191	addl s1(CTX,RT2,4),	RT0d; \
192	xorl s2(CTX,RT1,4),	RT0d; \
193	addl s3(CTX,RT3,4),	RT0d; \
194	xorq RT0,		x;
195
196#define add_preloaded_roundkey4() \
197	xorq RKEY,		RX0; \
198	xorq RKEY,		RX1; \
199	xorq RKEY,		RX2; \
200	xorq RKEY,		RX3;
201
202#define preload_roundkey_enc(n) \
203	movq p+4*(n)(CTX),	RKEY;
204
205#define add_roundkey_enc4(n) \
206	add_preloaded_roundkey4(); \
207	preload_roundkey_enc(n + 2);
208
209#define round_enc4(n) \
210	add_roundkey_enc4(n); \
211	\
212	F4(RX0); \
213	F4(RX1); \
214	F4(RX2); \
215	F4(RX3); \
216	\
217	F4(RX0); \
218	F4(RX1); \
219	F4(RX2); \
220	F4(RX3);
221
222#define preload_roundkey_dec(n) \
223	movq p+4*((n)-1)(CTX),	RKEY; \
224	rorq $32,		RKEY;
225
226#define add_roundkey_dec4(n) \
227	add_preloaded_roundkey4(); \
228	preload_roundkey_dec(n - 2);
229
230#define round_dec4(n) \
231	add_roundkey_dec4(n); \
232	\
233	F4(RX0); \
234	F4(RX1); \
235	F4(RX2); \
236	F4(RX3); \
237	\
238	F4(RX0); \
239	F4(RX1); \
240	F4(RX2); \
241	F4(RX3);
242
243#define read_block4() \
244	movq (RIO),		RX0; \
245	rorq $32,		RX0; \
246	bswapq 			RX0; \
247	\
248	movq 8(RIO),		RX1; \
249	rorq $32,		RX1; \
250	bswapq 			RX1; \
251	\
252	movq 16(RIO),		RX2; \
253	rorq $32,		RX2; \
254	bswapq 			RX2; \
255	\
256	movq 24(RIO),		RX3; \
257	rorq $32,		RX3; \
258	bswapq 			RX3;
259
260#define write_block4() \
261	bswapq 			RX0; \
262	movq RX0,		(RIO); \
263	\
264	bswapq 			RX1; \
265	movq RX1,		8(RIO); \
266	\
267	bswapq 			RX2; \
268	movq RX2,		16(RIO); \
269	\
270	bswapq 			RX3; \
271	movq RX3,		24(RIO);
272
273#define xor_block4() \
274	bswapq 			RX0; \
275	xorq RX0,		(RIO); \
276	\
277	bswapq 			RX1; \
278	xorq RX1,		8(RIO); \
279	\
280	bswapq 			RX2; \
281	xorq RX2,		16(RIO); \
282	\
283	bswapq 			RX3; \
284	xorq RX3,		24(RIO);
285
286ENTRY(__blowfish_enc_blk_4way)
287	/* input:
288	 *	%rdi: ctx
289	 *	%rsi: dst
290	 *	%rdx: src
291	 *	%rcx: bool, if true: xor output
292	 */
293	pushq %r12;
294	pushq %rbx;
295	pushq %rcx;
296
297	movq %rdi, CTX
298	movq %rsi, %r11;
299	movq %rdx, RIO;
300
301	preload_roundkey_enc(0);
302
303	read_block4();
304
305	round_enc4(0);
306	round_enc4(2);
307	round_enc4(4);
308	round_enc4(6);
309	round_enc4(8);
310	round_enc4(10);
311	round_enc4(12);
312	round_enc4(14);
313	add_preloaded_roundkey4();
314
315	popq %r12;
316	movq %r11, RIO;
317
318	test %r12b, %r12b;
319	jnz .L__enc_xor4;
320
321	write_block4();
322
323	popq %rbx;
324	popq %r12;
325	ret;
326
327.L__enc_xor4:
328	xor_block4();
329
330	popq %rbx;
331	popq %r12;
332	ret;
333ENDPROC(__blowfish_enc_blk_4way)
334
335ENTRY(blowfish_dec_blk_4way)
336	/* input:
337	 *	%rdi: ctx
338	 *	%rsi: dst
339	 *	%rdx: src
340	 */
341	pushq %r12;
342	pushq %rbx;
343
344	movq %rdi, CTX;
345	movq %rsi, %r11
346	movq %rdx, RIO;
347
348	preload_roundkey_dec(17);
349	read_block4();
350
351	round_dec4(17);
352	round_dec4(15);
353	round_dec4(13);
354	round_dec4(11);
355	round_dec4(9);
356	round_dec4(7);
357	round_dec4(5);
358	round_dec4(3);
359	add_preloaded_roundkey4();
360
361	movq %r11, RIO;
362	write_block4();
363
364	popq %rbx;
365	popq %r12;
366
367	ret;
368ENDPROC(blowfish_dec_blk_4way)
369