summaryrefslogtreecommitdiffstats
path: root/arch/s390/crypto/crc32le-vx.S
blob: 17f2504c26334472be0d562c1ef38344f197075f (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
/*
 * Hardware-accelerated CRC-32 variants for Linux on z Systems
 *
 * Use the z/Architecture Vector Extension Facility to accelerate the
 * computing of bitreflected CRC-32 checksums for IEEE 802.3 Ethernet
 * and Castagnoli.
 *
 * This CRC-32 implementation algorithm is bitreflected and processes
 * the least-significant bit first (Little-Endian).
 *
 * Copyright IBM Corp. 2015
 * Author(s): Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
 */

#include <linux/linkage.h>
#include <asm/vx-insn.h>

/* Vector register range containing CRC-32 constants */
#define CONST_PERM_LE2BE	%v9
#define CONST_R2R1		%v10
#define CONST_R4R3		%v11
#define CONST_R5		%v12
#define CONST_RU_POLY		%v13
#define CONST_CRC_POLY		%v14

.data
.align 8

/*
 * The CRC-32 constant block contains reduction constants to fold and
 * process particular chunks of the input data stream in parallel.
 *
 * For the CRC-32 variants, the constants are precomputed according to
 * these definitions:
 *
 *	R1 = [(x4*128+32 mod P'(x) << 32)]' << 1
 *	R2 = [(x4*128-32 mod P'(x) << 32)]' << 1
 *	R3 = [(x128+32 mod P'(x) << 32)]'   << 1
 *	R4 = [(x128-32 mod P'(x) << 32)]'   << 1
 *	R5 = [(x64 mod P'(x) << 32)]'	    << 1
 *	R6 = [(x32 mod P'(x) << 32)]'	    << 1
 *
 *	The bitreflected Barret reduction constant, u', is defined as
 *	the bit reversal of floor(x**64 / P(x)).
 *
 *	where P(x) is the polynomial in the normal domain and the P'(x) is the
 *	polynomial in the reversed (bitreflected) domain.
 *
 * CRC-32 (IEEE 802.3 Ethernet, ...) polynomials:
 *
 *	P(x)  = 0x04C11DB7
 *	P'(x) = 0xEDB88320
 *
 * CRC-32C (Castagnoli) polynomials:
 *
 *	P(x)  = 0x1EDC6F41
 *	P'(x) = 0x82F63B78
 */

.Lconstants_CRC_32_LE:
	.octa		0x0F0E0D0C0B0A09080706050403020100	# BE->LE mask
	.quad		0x1c6e41596, 0x154442bd4		# R2, R1
	.quad		0x0ccaa009e, 0x1751997d0		# R4, R3
	.octa		0x163cd6124				# R5
	.octa		0x1F7011641				# u'
	.octa		0x1DB710641				# P'(x) << 1

.Lconstants_CRC_32C_LE:
	.octa		0x0F0E0D0C0B0A09080706050403020100	# BE->LE mask
	.quad		0x09e4addf8, 0x740eef02			# R2, R1
	.quad		0x14cd00bd6, 0xf20c0dfe			# R4, R3
	.octa		0x0dd45aab8				# R5
	.octa		0x0dea713f1				# u'
	.octa		0x105ec76f0				# P'(x) << 1

.previous


.text

/*
 * The CRC-32 functions use these calling conventions:
 *
 * Parameters:
 *
 *	%r2:	Initial CRC value, typically ~0; and final CRC (return) value.
 *	%r3:	Input buffer pointer, performance might be improved if the
 *		buffer is on a doubleword boundary.
 *	%r4:	Length of the buffer, must be 64 bytes or greater.
 *
 * Register usage:
 *
 *	%r5:	CRC-32 constant pool base pointer.
 *	V0:	Initial CRC value and intermediate constants and results.
 *	V1..V4:	Data for CRC computation.
 *	V5..V8:	Next data chunks that are fetched from the input buffer.
 *	V9:	Constant for BE->LE conversion and shift operations
 *
 *	V10..V14: CRC-32 constants.
 */

ENTRY(crc32_le_vgfm_16)
	larl	%r5,.Lconstants_CRC_32_LE
	j	crc32_le_vgfm_generic

ENTRY(crc32c_le_vgfm_16)
	larl	%r5,.Lconstants_CRC_32C_LE
	j	crc32_le_vgfm_generic


crc32_le_vgfm_generic:
	/* Load CRC-32 constants */
	VLM	CONST_PERM_LE2BE,CONST_CRC_POLY,0,%r5

	/*
	 * Load the initial CRC value.
	 *
	 * The CRC value is loaded into the rightmost word of the
	 * vector register and is later XORed with the LSB portion
	 * of the loaded input data.
	 */
	VZERO	%v0			/* Clear V0 */
	VLVGF	%v0,%r2,3		/* Load CRC into rightmost word */

	/* Load a 64-byte data chunk and XOR with CRC */
	VLM	%v1,%v4,0,%r3		/* 64-bytes into V1..V4 */
	VPERM	%v1,%v1,%v1,CONST_PERM_LE2BE
	VPERM	%v2,%v2,%v2,CONST_PERM_LE2BE
	VPERM	%v3,%v3,%v3,CONST_PERM_LE2BE
	VPERM	%v4,%v4,%v4,CONST_PERM_LE2BE

	VX	%v1,%v0,%v1		/* V1 ^= CRC */
	aghi	%r3,64			/* BUF = BUF + 64 */
	aghi	%r4,-64			/* LEN = LEN - 64 */

	cghi	%r4,64
	jl	.Lless_than_64bytes

.Lfold_64bytes_loop:
	/* Load the next 64-byte data chunk into V5 to V8 */
	VLM	%v5,%v8,0,%r3
	VPERM	%v5,%v5,%v5,CONST_PERM_LE2BE
	VPERM	%v6,%v6,%v6,CONST_PERM_LE2BE
	VPERM	%v7,%v7,%v7,CONST_PERM_LE2BE
	VPERM	%v8,%v8,%v8,CONST_PERM_LE2BE

	/*
	 * Perform a GF(2) multiplication of the doublewords in V1 with
	 * the R1 and R2 reduction constants in V0.  The intermediate result
	 * is then folded (accumulated) with the next data chunk in V5 and
	 * stored in V1. Repeat this step for the register contents
	 * in V2, V3, and V4 respectively.
	 */
	VGFMAG	%v1,CONST_R2R1,%v1,%v5
	VGFMAG	%v2,CONST_R2R1,%v2,%v6
	VGFMAG	%v3,CONST_R2R1,%v3,%v7
	VGFMAG	%v4,CONST_R2R1,%v4,%v8

	aghi	%r3,64			/* BUF = BUF + 64 */
	aghi	%r4,-64			/* LEN = LEN - 64 */

	cghi	%r4,64
	jnl	.Lfold_64bytes_loop

.Lless_than_64bytes:
	/*
	 * Fold V1 to V4 into a single 128-bit value in V1.  Multiply V1 with R3
	 * and R4 and accumulating the next 128-bit chunk until a single 128-bit
	 * value remains.
	 */
	VGFMAG	%v1,CONST_R4R3,%v1,%v2
	VGFMAG	%v1,CONST_R4R3,%v1,%v3
	VGFMAG	%v1,CONST_R4R3,%v1,%v4

	cghi	%r4,16
	jl	.Lfinal_fold

.Lfold_16bytes_loop:

	VL	%v2,0,,%r3		/* Load next data chunk */
	VPERM	%v2,%v2,%v2,CONST_PERM_LE2BE
	VGFMAG	%v1,CONST_R4R3,%v1,%v2	/* Fold next data chunk */

	aghi	%r3,16
	aghi	%r4,-16

	cghi	%r4,16
	jnl	.Lfold_16bytes_loop

.Lfinal_fold:
	/*
	 * Set up a vector register for byte shifts.  The shift value must
	 * be loaded in bits 1-4 in byte element 7 of a vector register.
	 * Shift by 8 bytes: 0x40
	 * Shift by 4 bytes: 0x20
	 */
	VLEIB	%v9,0x40,7

	/*
	 * Prepare V0 for the next GF(2) multiplication: shift V0 by 8 bytes
	 * to move R4 into the rightmost doubleword and set the leftmost
	 * doubleword to 0x1.
	 */
	VSRLB	%v0,CONST_R4R3,%v9
	VLEIG	%v0,1,0

	/*
	 * Compute GF(2) product of V1 and V0.	The rightmost doubleword
	 * of V1 is multiplied with R4.  The leftmost doubleword of V1 is
	 * multiplied by 0x1 and is then XORed with rightmost product.
	 * Implicitly, the intermediate leftmost product becomes padded
	 */
	VGFMG	%v1,%v0,%v1

	/*
	 * Now do the final 32-bit fold by multiplying the rightmost word
	 * in V1 with R5 and XOR the result with the remaining bits in V1.
	 *
	 * To achieve this by a single VGFMAG, right shift V1 by a word
	 * and store the result in V2 which is then accumulated.  Use the
	 * vector unpack instruction to load the rightmost half of the
	 * doubleword into the rightmost doubleword element of V1; the other
	 * half is loaded in the leftmost doubleword.
	 * The vector register with CONST_R5 contains the R5 constant in the
	 * rightmost doubleword and the leftmost doubleword is zero to ignore
	 * the leftmost product of V1.
	 */
	VLEIB	%v9,0x20,7		  /* Shift by words */
	VSRLB	%v2,%v1,%v9		  /* Store remaining bits in V2 */
	VUPLLF	%v1,%v1			  /* Split rightmost doubleword */
	VGFMAG	%v1,CONST_R5,%v1,%v2	  /* V1 = (V1 * R5) XOR V2 */

	/*
	 * Apply a Barret reduction to compute the final 32-bit CRC value.
	 *
	 * The input values to the Barret reduction are the degree-63 polynomial
	 * in V1 (R(x)), degree-32 generator polynomial, and the reduction
	 * constant u.	The Barret reduction result is the CRC value of R(x) mod
	 * P(x).
	 *
	 * The Barret reduction algorithm is defined as:
	 *
	 *    1. T1(x) = floor( R(x) / x^32 ) GF2MUL u
	 *    2. T2(x) = floor( T1(x) / x^32 ) GF2MUL P(x)
	 *    3. C(x)  = R(x) XOR T2(x) mod x^32
	 *
	 *  Note: The leftmost doubleword of vector register containing
	 *  CONST_RU_POLY is zero and, thus, the intermediate GF(2) product
	 *  is zero and does not contribute to the final result.
	 */

	/* T1(x) = floor( R(x) / x^32 ) GF2MUL u */
	VUPLLF	%v2,%v1
	VGFMG	%v2,CONST_RU_POLY,%v2

	/*
	 * Compute the GF(2) product of the CRC polynomial with T1(x) in
	 * V2 and XOR the intermediate result, T2(x), with the value in V1.
	 * The final result is stored in word element 2 of V2.
	 */
	VUPLLF	%v2,%v2
	VGFMAG	%v2,CONST_CRC_POLY,%v2,%v1

.Ldone:
	VLGVF	%r2,%v2,2
	br	%r14

.previous