Richard Kuo | 075a46a | 2011-10-31 18:38:04 -0500 | [diff] [blame] | 1 | /* |
| 2 | * Checksum functions for Hexagon |
| 3 | * |
Richard Kuo | e1858b2 | 2012-09-19 16:22:02 -0500 | [diff] [blame] | 4 | * Copyright (c) 2010-2011, The Linux Foundation. All rights reserved. |
Richard Kuo | 075a46a | 2011-10-31 18:38:04 -0500 | [diff] [blame] | 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 | */ |
Alexander Duyck | 01cfbad | 2016-03-11 14:05:34 -0800 | [diff] [blame] | 63 | __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr, |
| 64 | __u32 len, __u8 proto, __wsum sum) |
Richard Kuo | 075a46a | 2011-10-31 18:38:04 -0500 | [diff] [blame] | 65 | { |
| 66 | return (__force __sum16)~from64to16( |
| 67 | (__force u64)saddr + (__force u64)daddr + |
| 68 | (__force u64)sum + ((len + proto) << 8)); |
| 69 | } |
| 70 | |
Alexander Duyck | 01cfbad | 2016-03-11 14:05:34 -0800 | [diff] [blame] | 71 | __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr, |
| 72 | __u32 len, __u8 proto, __wsum sum) |
Richard Kuo | 075a46a | 2011-10-31 18:38:04 -0500 | [diff] [blame] | 73 | { |
| 74 | u64 result; |
| 75 | |
| 76 | result = (__force u64)saddr + (__force u64)daddr + |
| 77 | (__force u64)sum + ((len + proto) << 8); |
| 78 | |
| 79 | /* Fold down to 32-bits so we don't lose in the typedef-less |
| 80 | network stack. */ |
| 81 | /* 64 to 33 */ |
| 82 | result = (result & 0xffffffffUL) + (result >> 32); |
| 83 | /* 33 to 32 */ |
| 84 | result = (result & 0xffffffffUL) + (result >> 32); |
| 85 | return (__force __wsum)result; |
| 86 | } |
| 87 | EXPORT_SYMBOL(csum_tcpudp_nofold); |
| 88 | |
| 89 | /* |
| 90 | * Do a 64-bit checksum on an arbitrary memory area.. |
| 91 | * |
| 92 | * This isn't a great routine, but it's not _horrible_ either. The |
| 93 | * inner loop could be unrolled a bit further, and there are better |
| 94 | * ways to do the carry, but this is reasonable. |
| 95 | */ |
| 96 | |
| 97 | /* optimized HEXAGON intrinsic version, with over read fixed */ |
| 98 | unsigned int do_csum(const void *voidptr, int len) |
| 99 | { |
| 100 | u64 sum0, sum1, x0, x1, *ptr8_o, *ptr8_e, *ptr8; |
| 101 | int i, start, mid, end, mask; |
| 102 | const char *ptr = voidptr; |
| 103 | unsigned short *ptr2; |
| 104 | unsigned int *ptr4; |
| 105 | |
| 106 | if (len <= 0) |
| 107 | return 0; |
| 108 | |
| 109 | start = 0xF & (16-(((int) ptr) & 0xF)) ; |
| 110 | mask = 0x7fffffffUL >> HEXAGON_R_cl0_R(len); |
| 111 | start = start & mask ; |
| 112 | |
| 113 | mid = len - start; |
| 114 | end = mid & 0xF; |
| 115 | mid = mid>>4; |
| 116 | sum0 = mid << 18; |
| 117 | sum1 = 0; |
| 118 | |
| 119 | if (start & 1) |
| 120 | sum0 += (u64) (ptr[0] << 8); |
| 121 | ptr2 = (unsigned short *) &ptr[start & 1]; |
| 122 | if (start & 2) |
| 123 | sum1 += (u64) ptr2[0]; |
| 124 | ptr4 = (unsigned int *) &ptr[start & 3]; |
| 125 | if (start & 4) { |
| 126 | sum0 = HEXAGON_P_vrmpyhacc_PP(sum0, |
| 127 | VR_NEGATE(0, 0, 1, 1)^((u64)ptr4[0]), |
| 128 | VR_SELECT(0, 0, 1, 1)); |
| 129 | sum0 += VR_SELECT(0, 0, 1, 0); |
| 130 | } |
| 131 | ptr8 = (u64 *) &ptr[start & 7]; |
| 132 | if (start & 8) { |
| 133 | sum1 = HEXAGON_P_vrmpyhacc_PP(sum1, |
| 134 | VR_NEGATE(1, 1, 1, 1)^(ptr8[0]), |
| 135 | VR_SELECT(1, 1, 1, 1)); |
| 136 | sum1 += VR_CARRY(0, 0, 1, 0); |
| 137 | } |
| 138 | ptr8_o = (u64 *) (ptr + start); |
| 139 | ptr8_e = (u64 *) (ptr + start + 8); |
| 140 | |
| 141 | if (mid) { |
| 142 | x0 = *ptr8_e; ptr8_e += 2; |
| 143 | x1 = *ptr8_o; ptr8_o += 2; |
| 144 | if (mid > 1) |
| 145 | for (i = 0; i < mid-1; i++) { |
| 146 | sum0 = HEXAGON_P_vrmpyhacc_PP(sum0, |
| 147 | x0^VR_NEGATE(1, 1, 1, 1), |
| 148 | VR_SELECT(1, 1, 1, 1)); |
| 149 | sum1 = HEXAGON_P_vrmpyhacc_PP(sum1, |
| 150 | x1^VR_NEGATE(1, 1, 1, 1), |
| 151 | VR_SELECT(1, 1, 1, 1)); |
| 152 | x0 = *ptr8_e; ptr8_e += 2; |
| 153 | x1 = *ptr8_o; ptr8_o += 2; |
| 154 | } |
| 155 | sum0 = HEXAGON_P_vrmpyhacc_PP(sum0, x0^VR_NEGATE(1, 1, 1, 1), |
| 156 | VR_SELECT(1, 1, 1, 1)); |
| 157 | sum1 = HEXAGON_P_vrmpyhacc_PP(sum1, x1^VR_NEGATE(1, 1, 1, 1), |
| 158 | VR_SELECT(1, 1, 1, 1)); |
| 159 | } |
| 160 | |
| 161 | ptr4 = (unsigned int *) &ptr[start + (mid * 16) + (end & 8)]; |
| 162 | if (end & 4) { |
| 163 | sum1 = HEXAGON_P_vrmpyhacc_PP(sum1, |
| 164 | VR_NEGATE(0, 0, 1, 1)^((u64)ptr4[0]), |
| 165 | VR_SELECT(0, 0, 1, 1)); |
| 166 | sum1 += VR_SELECT(0, 0, 1, 0); |
| 167 | } |
| 168 | ptr2 = (unsigned short *) &ptr[start + (mid * 16) + (end & 12)]; |
| 169 | if (end & 2) |
| 170 | sum0 += (u64) ptr2[0]; |
| 171 | |
| 172 | if (end & 1) |
| 173 | sum1 += (u64) ptr[start + (mid * 16) + (end & 14)]; |
| 174 | |
| 175 | ptr8 = (u64 *) &ptr[start + (mid * 16)]; |
| 176 | if (end & 8) { |
| 177 | sum0 = HEXAGON_P_vrmpyhacc_PP(sum0, |
| 178 | VR_NEGATE(1, 1, 1, 1)^(ptr8[0]), |
| 179 | VR_SELECT(1, 1, 1, 1)); |
| 180 | sum0 += VR_CARRY(0, 0, 1, 0); |
| 181 | } |
| 182 | sum0 = HEXAGON_P_vrmpyh_PP((sum0+sum1)^VR_NEGATE(0, 0, 0, 1), |
| 183 | VR_SELECT(0, 0, 1, 1)); |
| 184 | sum0 += VR_NEGATE(0, 0, 0, 1); |
| 185 | sum0 = HEXAGON_P_vrmpyh_PP(sum0, VR_SELECT(0, 0, 1, 1)); |
| 186 | |
| 187 | if (start & 1) |
| 188 | sum0 = (sum0 << 8) | (0xFF & (sum0 >> 8)); |
| 189 | |
| 190 | return 0xFFFF & sum0; |
| 191 | } |
| 192 | |
| 193 | /* |
| 194 | * copy from ds while checksumming, otherwise like csum_partial |
| 195 | */ |
| 196 | __wsum |
| 197 | csum_partial_copy_nocheck(const void *src, void *dst, int len, __wsum sum) |
| 198 | { |
| 199 | memcpy(dst, src, len); |
| 200 | return csum_partial(dst, len, sum); |
| 201 | } |