1/* GPL HEADER START
2 *
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
12 * General Public License version 2 for more details (a copy is included
13 * in the LICENSE file that accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License
16 * version 2 along with this program; If not, see http://www.gnu.org/licenses
17 *
18 * Please  visit http://www.xyratex.com/contact if you need additional
19 * information or have any questions.
20 *
21 * GPL HEADER END
22 */
23
24/*
25 * Copyright 2012 Xyratex Technology Limited
26 *
27 * Using hardware provided PCLMULQDQ instruction to accelerate the CRC32
28 * calculation.
29 * CRC32 polynomial:0x04c11db7(BE)/0xEDB88320(LE)
30 * PCLMULQDQ is a new instruction in Intel SSE4.2, the reference can be found
31 * at:
32 * http://www.intel.com/products/processor/manuals/
33 * Intel(R) 64 and IA-32 Architectures Software Developer's Manual
34 * Volume 2B: Instruction Set Reference, N-Z
35 *
36 * Authors:   Gregory Prestas <Gregory_Prestas@us.xyratex.com>
37 *	      Alexander Boyko <Alexander_Boyko@xyratex.com>
38 */
39
40#include <linux/linkage.h>
41
42
43.section .rodata
44.align 16
45/*
46 * [x4*128+32 mod P(x) << 32)]'  << 1   = 0x154442bd4
47 * #define CONSTANT_R1  0x154442bd4LL
48 *
49 * [(x4*128-32 mod P(x) << 32)]' << 1   = 0x1c6e41596
50 * #define CONSTANT_R2  0x1c6e41596LL
51 */
52.Lconstant_R2R1:
53	.octa 0x00000001c6e415960000000154442bd4
54/*
55 * [(x128+32 mod P(x) << 32)]'   << 1   = 0x1751997d0
56 * #define CONSTANT_R3  0x1751997d0LL
57 *
58 * [(x128-32 mod P(x) << 32)]'   << 1   = 0x0ccaa009e
59 * #define CONSTANT_R4  0x0ccaa009eLL
60 */
61.Lconstant_R4R3:
62	.octa 0x00000000ccaa009e00000001751997d0
63/*
64 * [(x64 mod P(x) << 32)]'       << 1   = 0x163cd6124
65 * #define CONSTANT_R5  0x163cd6124LL
66 */
67.Lconstant_R5:
68	.octa 0x00000000000000000000000163cd6124
69.Lconstant_mask32:
70	.octa 0x000000000000000000000000FFFFFFFF
71/*
72 * #define CRCPOLY_TRUE_LE_FULL 0x1DB710641LL
73 *
74 * Barrett Reduction constant (u64`) = u` = (x**64 / P(x))` = 0x1F7011641LL
75 * #define CONSTANT_RU  0x1F7011641LL
76 */
77.Lconstant_RUpoly:
78	.octa 0x00000001F701164100000001DB710641
79
80#define CONSTANT %xmm0
81
82#ifdef __x86_64__
83#define BUF     %rdi
84#define LEN     %rsi
85#define CRC     %edx
86#else
87#define BUF     %eax
88#define LEN     %edx
89#define CRC     %ecx
90#endif
91
92
93
94.text
95/**
96 *      Calculate crc32
97 *      BUF - buffer (16 bytes aligned)
98 *      LEN - sizeof buffer (16 bytes aligned), LEN should be grater than 63
99 *      CRC - initial crc32
100 *      return %eax crc32
101 *      uint crc32_pclmul_le_16(unsigned char const *buffer,
102 *	                     size_t len, uint crc32)
103 */
104
105SYM_FUNC_START(crc32_pclmul_le_16) /* buffer and buffer size are 16 bytes aligned */
106	movdqa  (BUF), %xmm1
107	movdqa  0x10(BUF), %xmm2
108	movdqa  0x20(BUF), %xmm3
109	movdqa  0x30(BUF), %xmm4
110	movd    CRC, CONSTANT
111	pxor    CONSTANT, %xmm1
112	sub     $0x40, LEN
113	add     $0x40, BUF
114	cmp     $0x40, LEN
115	jb      less_64
116
117#ifdef __x86_64__
118	movdqa .Lconstant_R2R1(%rip), CONSTANT
119#else
120	movdqa .Lconstant_R2R1, CONSTANT
121#endif
122
123loop_64:/*  64 bytes Full cache line folding */
124	prefetchnta    0x40(BUF)
125	movdqa  %xmm1, %xmm5
126	movdqa  %xmm2, %xmm6
127	movdqa  %xmm3, %xmm7
128#ifdef __x86_64__
129	movdqa  %xmm4, %xmm8
130#endif
131	pclmulqdq $0x00, CONSTANT, %xmm1
132	pclmulqdq $0x00, CONSTANT, %xmm2
133	pclmulqdq $0x00, CONSTANT, %xmm3
134#ifdef __x86_64__
135	pclmulqdq $0x00, CONSTANT, %xmm4
136#endif
137	pclmulqdq $0x11, CONSTANT, %xmm5
138	pclmulqdq $0x11, CONSTANT, %xmm6
139	pclmulqdq $0x11, CONSTANT, %xmm7
140#ifdef __x86_64__
141	pclmulqdq $0x11, CONSTANT, %xmm8
142#endif
143	pxor    %xmm5, %xmm1
144	pxor    %xmm6, %xmm2
145	pxor    %xmm7, %xmm3
146#ifdef __x86_64__
147	pxor    %xmm8, %xmm4
148#else
149	/* xmm8 unsupported for x32 */
150	movdqa  %xmm4, %xmm5
151	pclmulqdq $0x00, CONSTANT, %xmm4
152	pclmulqdq $0x11, CONSTANT, %xmm5
153	pxor    %xmm5, %xmm4
154#endif
155
156	pxor    (BUF), %xmm1
157	pxor    0x10(BUF), %xmm2
158	pxor    0x20(BUF), %xmm3
159	pxor    0x30(BUF), %xmm4
160
161	sub     $0x40, LEN
162	add     $0x40, BUF
163	cmp     $0x40, LEN
164	jge     loop_64
165less_64:/*  Folding cache line into 128bit */
166#ifdef __x86_64__
167	movdqa  .Lconstant_R4R3(%rip), CONSTANT
168#else
169	movdqa  .Lconstant_R4R3, CONSTANT
170#endif
171	prefetchnta     (BUF)
172
173	movdqa  %xmm1, %xmm5
174	pclmulqdq $0x00, CONSTANT, %xmm1
175	pclmulqdq $0x11, CONSTANT, %xmm5
176	pxor    %xmm5, %xmm1
177	pxor    %xmm2, %xmm1
178
179	movdqa  %xmm1, %xmm5
180	pclmulqdq $0x00, CONSTANT, %xmm1
181	pclmulqdq $0x11, CONSTANT, %xmm5
182	pxor    %xmm5, %xmm1
183	pxor    %xmm3, %xmm1
184
185	movdqa  %xmm1, %xmm5
186	pclmulqdq $0x00, CONSTANT, %xmm1
187	pclmulqdq $0x11, CONSTANT, %xmm5
188	pxor    %xmm5, %xmm1
189	pxor    %xmm4, %xmm1
190
191	cmp     $0x10, LEN
192	jb      fold_64
193loop_16:/* Folding rest buffer into 128bit */
194	movdqa  %xmm1, %xmm5
195	pclmulqdq $0x00, CONSTANT, %xmm1
196	pclmulqdq $0x11, CONSTANT, %xmm5
197	pxor    %xmm5, %xmm1
198	pxor    (BUF), %xmm1
199	sub     $0x10, LEN
200	add     $0x10, BUF
201	cmp     $0x10, LEN
202	jge     loop_16
203
204fold_64:
205	/* perform the last 64 bit fold, also adds 32 zeroes
206	 * to the input stream */
207	pclmulqdq $0x01, %xmm1, CONSTANT /* R4 * xmm1.low */
208	psrldq  $0x08, %xmm1
209	pxor    CONSTANT, %xmm1
210
211	/* final 32-bit fold */
212	movdqa  %xmm1, %xmm2
213#ifdef __x86_64__
214	movdqa  .Lconstant_R5(%rip), CONSTANT
215	movdqa  .Lconstant_mask32(%rip), %xmm3
216#else
217	movdqa  .Lconstant_R5, CONSTANT
218	movdqa  .Lconstant_mask32, %xmm3
219#endif
220	psrldq  $0x04, %xmm2
221	pand    %xmm3, %xmm1
222	pclmulqdq $0x00, CONSTANT, %xmm1
223	pxor    %xmm2, %xmm1
224
225	/* Finish up with the bit-reversed barrett reduction 64 ==> 32 bits */
226#ifdef __x86_64__
227	movdqa  .Lconstant_RUpoly(%rip), CONSTANT
228#else
229	movdqa  .Lconstant_RUpoly, CONSTANT
230#endif
231	movdqa  %xmm1, %xmm2
232	pand    %xmm3, %xmm1
233	pclmulqdq $0x10, CONSTANT, %xmm1
234	pand    %xmm3, %xmm1
235	pclmulqdq $0x00, CONSTANT, %xmm1
236	pxor    %xmm2, %xmm1
237	pextrd  $0x01, %xmm1, %eax
238
239	RET
240SYM_FUNC_END(crc32_pclmul_le_16)
241