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
102SYM_FUNC_START(blowfish_enc_blk)
103	/* input:
104	 *	%rdi: ctx
105	 *	%rsi: dst
106	 *	%rdx: src
107	 */
108	movq %r12, %r11;
109
110	movq %rdi, CTX;
111	movq %rsi, %r10;
112	movq %rdx, RIO;
113
114	read_block();
115
116	round_enc(0);
117	round_enc(2);
118	round_enc(4);
119	round_enc(6);
120	round_enc(8);
121	round_enc(10);
122	round_enc(12);
123	round_enc(14);
124	add_roundkey_enc(16);
125
126	movq %r11, %r12;
127	movq %r10, RIO;
128
129	write_block();
130	RET;
131SYM_FUNC_END(blowfish_enc_blk)
132
133SYM_FUNC_START(blowfish_dec_blk)
134	/* input:
135	 *	%rdi: ctx
136	 *	%rsi: dst
137	 *	%rdx: src
138	 */
139	movq %r12, %r11;
140
141	movq %rdi, CTX;
142	movq %rsi, %r10;
143	movq %rdx, RIO;
144
145	read_block();
146
147	round_dec(17);
148	round_dec(15);
149	round_dec(13);
150	round_dec(11);
151	round_dec(9);
152	round_dec(7);
153	round_dec(5);
154	round_dec(3);
155	add_roundkey_dec(1);
156
157	movq %r10, RIO;
158	write_block();
159
160	movq %r11, %r12;
161
162	RET;
163SYM_FUNC_END(blowfish_dec_blk)
164
165/**********************************************************************
166  4-way blowfish, four blocks parallel
167 **********************************************************************/
168
169/* F() for 4-way. Slower when used alone/1-way, but faster when used
170 * parallel/4-way (tested on AMD Phenom II & Intel Xeon E7330).
171 */
172#define F4(x) \
173	movzbl x ## bh,		RT1d; \
174	movzbl x ## bl,		RT3d; \
175	rorq $16,		x; \
176	movzbl x ## bh,		RT0d; \
177	movzbl x ## bl,		RT2d; \
178	rorq $16,		x; \
179	movl s0(CTX,RT0,4),	RT0d; \
180	addl s1(CTX,RT2,4),	RT0d; \
181	xorl s2(CTX,RT1,4),	RT0d; \
182	addl s3(CTX,RT3,4),	RT0d; \
183	xorq RT0,		x;
184
185#define add_preloaded_roundkey4() \
186	xorq RKEY,		RX0; \
187	xorq RKEY,		RX1; \
188	xorq RKEY,		RX2; \
189	xorq RKEY,		RX3;
190
191#define preload_roundkey_enc(n) \
192	movq p+4*(n)(CTX),	RKEY;
193
194#define add_roundkey_enc4(n) \
195	add_preloaded_roundkey4(); \
196	preload_roundkey_enc(n + 2);
197
198#define round_enc4(n) \
199	add_roundkey_enc4(n); \
200	\
201	F4(RX0); \
202	F4(RX1); \
203	F4(RX2); \
204	F4(RX3); \
205	\
206	F4(RX0); \
207	F4(RX1); \
208	F4(RX2); \
209	F4(RX3);
210
211#define preload_roundkey_dec(n) \
212	movq p+4*((n)-1)(CTX),	RKEY; \
213	rorq $32,		RKEY;
214
215#define add_roundkey_dec4(n) \
216	add_preloaded_roundkey4(); \
217	preload_roundkey_dec(n - 2);
218
219#define round_dec4(n) \
220	add_roundkey_dec4(n); \
221	\
222	F4(RX0); \
223	F4(RX1); \
224	F4(RX2); \
225	F4(RX3); \
226	\
227	F4(RX0); \
228	F4(RX1); \
229	F4(RX2); \
230	F4(RX3);
231
232#define read_block4() \
233	movq (RIO),		RX0; \
234	rorq $32,		RX0; \
235	bswapq 			RX0; \
236	\
237	movq 8(RIO),		RX1; \
238	rorq $32,		RX1; \
239	bswapq 			RX1; \
240	\
241	movq 16(RIO),		RX2; \
242	rorq $32,		RX2; \
243	bswapq 			RX2; \
244	\
245	movq 24(RIO),		RX3; \
246	rorq $32,		RX3; \
247	bswapq 			RX3;
248
249#define write_block4() \
250	bswapq 			RX0; \
251	movq RX0,		(RIO); \
252	\
253	bswapq 			RX1; \
254	movq RX1,		8(RIO); \
255	\
256	bswapq 			RX2; \
257	movq RX2,		16(RIO); \
258	\
259	bswapq 			RX3; \
260	movq RX3,		24(RIO);
261
262#define xor_block4() \
263	movq (RIO),		RT0; \
264	bswapq			RT0; \
265	xorq RT0,		RX1; \
266	\
267	movq 8(RIO),		RT2; \
268	bswapq			RT2; \
269	xorq RT2,		RX2; \
270	\
271	movq 16(RIO),		RT3; \
272	bswapq			RT3; \
273	xorq RT3,		RX3;
274
275SYM_FUNC_START(blowfish_enc_blk_4way)
276	/* input:
277	 *	%rdi: ctx
278	 *	%rsi: dst
279	 *	%rdx: src
280	 */
281	pushq %r12;
282	pushq %rbx;
283
284	movq %rdi, CTX
285	movq %rsi, %r11;
286	movq %rdx, RIO;
287
288	preload_roundkey_enc(0);
289
290	read_block4();
291
292	round_enc4(0);
293	round_enc4(2);
294	round_enc4(4);
295	round_enc4(6);
296	round_enc4(8);
297	round_enc4(10);
298	round_enc4(12);
299	round_enc4(14);
300	add_preloaded_roundkey4();
301
302	movq %r11, RIO;
303	write_block4();
304
305	popq %rbx;
306	popq %r12;
307	RET;
308SYM_FUNC_END(blowfish_enc_blk_4way)
309
310SYM_FUNC_START(__blowfish_dec_blk_4way)
311	/* input:
312	 *	%rdi: ctx
313	 *	%rsi: dst
314	 *	%rdx: src
315	 *	%rcx: cbc (bool)
316	 */
317	pushq %r12;
318	pushq %rbx;
319	pushq %rcx;
320	pushq %rdx;
321
322	movq %rdi, CTX;
323	movq %rsi, %r11;
324	movq %rdx, RIO;
325
326	preload_roundkey_dec(17);
327	read_block4();
328
329	round_dec4(17);
330	round_dec4(15);
331	round_dec4(13);
332	round_dec4(11);
333	round_dec4(9);
334	round_dec4(7);
335	round_dec4(5);
336	round_dec4(3);
337	add_preloaded_roundkey4();
338
339	popq RIO;
340	popq %r12;
341	testq %r12, %r12;
342	jz .L_no_cbc_xor;
343
344	xor_block4();
345
346.L_no_cbc_xor:
347	movq %r11, RIO;
348	write_block4();
349
350	popq %rbx;
351	popq %r12;
352
353	RET;
354SYM_FUNC_END(__blowfish_dec_blk_4way)
355