xref: /openbmc/linux/arch/hexagon/lib/checksum.c (revision 588b48ca)
1 /*
2  * Checksum functions for Hexagon
3  *
4  * Copyright (c) 2010-2011, The Linux Foundation. All rights reserved.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 and
8  * only version 2 as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write to the Free Software
17  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
18  * 02110-1301, USA.
19  */
20 
21 /*  This was derived from arch/alpha/lib/checksum.c  */
22 
23 
24 #include <linux/module.h>
25 #include <linux/string.h>
26 
27 #include <asm/byteorder.h>
28 #include <net/checksum.h>
29 #include <linux/uaccess.h>
30 #include <asm/intrinsics.h>
31 
32 
33 /*  Vector value operations  */
34 #define SIGN(x, y)	((0x8000ULL*x)<<y)
35 #define CARRY(x, y)	((0x0002ULL*x)<<y)
36 #define SELECT(x, y)	((0x0001ULL*x)<<y)
37 
38 #define VR_NEGATE(a, b, c, d)	(SIGN(a, 48) + SIGN(b, 32) + SIGN(c, 16) \
39 	+ SIGN(d, 0))
40 #define VR_CARRY(a, b, c, d)	(CARRY(a, 48) + CARRY(b, 32) + CARRY(c, 16) \
41 	+ CARRY(d, 0))
42 #define VR_SELECT(a, b, c, d)	(SELECT(a, 48) + SELECT(b, 32) + SELECT(c, 16) \
43 	+ SELECT(d, 0))
44 
45 
46 /* optimized HEXAGON V3 intrinsic version */
47 static inline unsigned short from64to16(u64 x)
48 {
49 	u64 sum;
50 
51 	sum = HEXAGON_P_vrmpyh_PP(x^VR_NEGATE(1, 1, 1, 1),
52 			     VR_SELECT(1, 1, 1, 1));
53 	sum += VR_CARRY(0, 0, 1, 0);
54 	sum = HEXAGON_P_vrmpyh_PP(sum, VR_SELECT(0, 0, 1, 1));
55 
56 	return 0xFFFF & sum;
57 }
58 
59 /*
60  * computes the checksum of the TCP/UDP pseudo-header
61  * returns a 16-bit checksum, already complemented.
62  */
63 __sum16 csum_tcpudp_magic(unsigned long saddr, unsigned long daddr,
64 			  unsigned short len, unsigned short proto,
65 			  __wsum sum)
66 {
67 	return (__force __sum16)~from64to16(
68 		(__force u64)saddr + (__force u64)daddr +
69 		(__force u64)sum + ((len + proto) << 8));
70 }
71 
72 __wsum csum_tcpudp_nofold(unsigned long saddr, unsigned long daddr,
73 			  unsigned short len, unsigned short proto,
74 			  __wsum sum)
75 {
76 	u64 result;
77 
78 	result = (__force u64)saddr + (__force u64)daddr +
79 		 (__force u64)sum + ((len + proto) << 8);
80 
81 	/* Fold down to 32-bits so we don't lose in the typedef-less
82 	   network stack.  */
83 	/* 64 to 33 */
84 	result = (result & 0xffffffffUL) + (result >> 32);
85 	/* 33 to 32 */
86 	result = (result & 0xffffffffUL) + (result >> 32);
87 	return (__force __wsum)result;
88 }
89 EXPORT_SYMBOL(csum_tcpudp_nofold);
90 
91 /*
92  * Do a 64-bit checksum on an arbitrary memory area..
93  *
94  * This isn't a great routine, but it's not _horrible_ either. The
95  * inner loop could be unrolled a bit further, and there are better
96  * ways to do the carry, but this is reasonable.
97  */
98 
99 /* optimized HEXAGON intrinsic version, with over read fixed */
100 unsigned int do_csum(const void *voidptr, int len)
101 {
102 	u64 sum0, sum1, x0, x1, *ptr8_o, *ptr8_e, *ptr8;
103 	int i, start, mid, end, mask;
104 	const char *ptr = voidptr;
105 	unsigned short *ptr2;
106 	unsigned int *ptr4;
107 
108 	if (len <= 0)
109 		return 0;
110 
111 	start = 0xF & (16-(((int) ptr) & 0xF)) ;
112 	mask  = 0x7fffffffUL >> HEXAGON_R_cl0_R(len);
113 	start = start & mask ;
114 
115 	mid = len - start;
116 	end = mid & 0xF;
117 	mid = mid>>4;
118 	sum0 = mid << 18;
119 	sum1 = 0;
120 
121 	if (start & 1)
122 		sum0 += (u64) (ptr[0] << 8);
123 	ptr2 = (unsigned short *) &ptr[start & 1];
124 	if (start & 2)
125 		sum1 += (u64) ptr2[0];
126 	ptr4 = (unsigned int *) &ptr[start & 3];
127 	if (start & 4) {
128 		sum0 = HEXAGON_P_vrmpyhacc_PP(sum0,
129 			VR_NEGATE(0, 0, 1, 1)^((u64)ptr4[0]),
130 			VR_SELECT(0, 0, 1, 1));
131 		sum0 += VR_SELECT(0, 0, 1, 0);
132 	}
133 	ptr8 = (u64 *) &ptr[start & 7];
134 	if (start & 8) {
135 		sum1 = HEXAGON_P_vrmpyhacc_PP(sum1,
136 			VR_NEGATE(1, 1, 1, 1)^(ptr8[0]),
137 			VR_SELECT(1, 1, 1, 1));
138 		sum1 += VR_CARRY(0, 0, 1, 0);
139 	}
140 	ptr8_o = (u64 *) (ptr + start);
141 	ptr8_e = (u64 *) (ptr + start + 8);
142 
143 	if (mid) {
144 		x0 = *ptr8_e; ptr8_e += 2;
145 		x1 = *ptr8_o; ptr8_o += 2;
146 		if (mid > 1)
147 			for (i = 0; i < mid-1; i++) {
148 				sum0 = HEXAGON_P_vrmpyhacc_PP(sum0,
149 					x0^VR_NEGATE(1, 1, 1, 1),
150 					VR_SELECT(1, 1, 1, 1));
151 				sum1 = HEXAGON_P_vrmpyhacc_PP(sum1,
152 					x1^VR_NEGATE(1, 1, 1, 1),
153 					VR_SELECT(1, 1, 1, 1));
154 				x0 = *ptr8_e; ptr8_e += 2;
155 				x1 = *ptr8_o; ptr8_o += 2;
156 			}
157 		sum0 = HEXAGON_P_vrmpyhacc_PP(sum0, x0^VR_NEGATE(1, 1, 1, 1),
158 			VR_SELECT(1, 1, 1, 1));
159 		sum1 = HEXAGON_P_vrmpyhacc_PP(sum1, x1^VR_NEGATE(1, 1, 1, 1),
160 			VR_SELECT(1, 1, 1, 1));
161 	}
162 
163 	ptr4 = (unsigned int *) &ptr[start + (mid * 16) + (end & 8)];
164 	if (end & 4) {
165 		sum1 = HEXAGON_P_vrmpyhacc_PP(sum1,
166 			VR_NEGATE(0, 0, 1, 1)^((u64)ptr4[0]),
167 			VR_SELECT(0, 0, 1, 1));
168 		sum1 += VR_SELECT(0, 0, 1, 0);
169 	}
170 	ptr2 = (unsigned short *) &ptr[start + (mid * 16) + (end & 12)];
171 	if (end & 2)
172 		sum0 += (u64) ptr2[0];
173 
174 	if (end & 1)
175 		sum1 += (u64) ptr[start + (mid * 16) + (end & 14)];
176 
177 	ptr8 = (u64 *) &ptr[start + (mid * 16)];
178 	if (end & 8) {
179 		sum0 = HEXAGON_P_vrmpyhacc_PP(sum0,
180 			VR_NEGATE(1, 1, 1, 1)^(ptr8[0]),
181 			VR_SELECT(1, 1, 1, 1));
182 		sum0 += VR_CARRY(0, 0, 1, 0);
183 	}
184 	sum0 = HEXAGON_P_vrmpyh_PP((sum0+sum1)^VR_NEGATE(0, 0, 0, 1),
185 		VR_SELECT(0, 0, 1, 1));
186 	sum0 += VR_NEGATE(0, 0, 0, 1);
187 	sum0 = HEXAGON_P_vrmpyh_PP(sum0, VR_SELECT(0, 0, 1, 1));
188 
189 	if (start & 1)
190 		sum0 = (sum0 << 8) | (0xFF & (sum0 >> 8));
191 
192 	return 0xFFFF & sum0;
193 }
194 
195 /*
196  * copy from ds while checksumming, otherwise like csum_partial
197  */
198 __wsum
199 csum_partial_copy_nocheck(const void *src, void *dst, int len, __wsum sum)
200 {
201 	memcpy(dst, src, len);
202 	return csum_partial(dst, len, sum);
203 }
204