Chen Liqin | 6bc9a39 | 2009-06-12 22:01:00 +0800 | [diff] [blame] | 1 | #ifndef _ASM_SCORE_CHECKSUM_H |
| 2 | #define _ASM_SCORE_CHECKSUM_H |
| 3 | |
| 4 | #include <linux/in6.h> |
| 5 | #include <asm/uaccess.h> |
| 6 | |
| 7 | /* |
| 8 | * computes the checksum of a memory block at buff, length len, |
| 9 | * and adds in "sum" (32-bit) |
| 10 | * |
| 11 | * returns a 32-bit number suitable for feeding into itself |
| 12 | * or csum_tcpudp_magic |
| 13 | * |
| 14 | * this function must be called with even lengths, except |
| 15 | * for the last fragment, which may be odd |
| 16 | * |
| 17 | * it's best to have buff aligned on a 32-bit boundary |
| 18 | */ |
| 19 | unsigned int csum_partial(const void *buff, int len, __wsum sum); |
| 20 | unsigned int csum_partial_copy_from_user(const char *src, char *dst, int len, |
| 21 | unsigned int sum, int *csum_err); |
| 22 | unsigned int csum_partial_copy(const char *src, char *dst, |
| 23 | int len, unsigned int sum); |
| 24 | |
| 25 | /* |
| 26 | * this is a new version of the above that records errors it finds in *errp, |
| 27 | * but continues and zeros the rest of the buffer. |
| 28 | */ |
| 29 | |
| 30 | /* |
| 31 | * Copy and checksum to user |
| 32 | */ |
| 33 | #define HAVE_CSUM_COPY_USER |
| 34 | static inline |
| 35 | __wsum csum_and_copy_to_user(const void *src, void __user *dst, int len, |
| 36 | __wsum sum, int *err_ptr) |
| 37 | { |
| 38 | sum = csum_partial(src, len, sum); |
| 39 | if (copy_to_user(dst, src, len)) { |
| 40 | *err_ptr = -EFAULT; |
| 41 | return (__force __wsum) -1; /* invalid checksum */ |
| 42 | } |
| 43 | return sum; |
| 44 | } |
| 45 | |
| 46 | |
| 47 | #define csum_partial_copy_nocheck csum_partial_copy |
| 48 | /* |
| 49 | * Fold a partial checksum without adding pseudo headers |
| 50 | */ |
| 51 | |
| 52 | static inline __sum16 csum_fold(__wsum sum) |
| 53 | { |
| 54 | /* the while loop is unnecessary really, it's always enough with two |
| 55 | iterations */ |
| 56 | __asm__ __volatile__( |
| 57 | ".set volatile\n\t" |
| 58 | ".set\tr1\n\t" |
| 59 | "slli\tr1,%0, 16\n\t" |
| 60 | "add\t%0,%0, r1\n\t" |
| 61 | "cmp.c\tr1, %0\n\t" |
| 62 | "srli\t%0, %0, 16\n\t" |
| 63 | "bleu\t1f\n\t" |
| 64 | "addi\t%0, 0x1\n\t" |
| 65 | "1:ldi\tr30, 0xffff\n\t" |
| 66 | "xor\t%0, %0, r30\n\t" |
| 67 | "slli\t%0, %0, 16\n\t" |
| 68 | "srli\t%0, %0, 16\n\t" |
| 69 | ".set\tnor1\n\t" |
| 70 | ".set optimize\n\t" |
| 71 | : "=r" (sum) |
| 72 | : "0" (sum)); |
| 73 | return sum; |
| 74 | } |
| 75 | |
| 76 | /* |
| 77 | * This is a version of ip_compute_csum() optimized for IP headers, |
| 78 | * which always checksum on 4 octet boundaries. |
| 79 | * |
| 80 | * By Jorge Cwik <jorge@laser.satlink.net>, adapted for linux by |
| 81 | * Arnt Gulbrandsen. |
| 82 | */ |
| 83 | static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl) |
| 84 | { |
| 85 | unsigned int sum; |
| 86 | unsigned long dummy; |
| 87 | |
| 88 | __asm__ __volatile__( |
| 89 | ".set volatile\n\t" |
| 90 | ".set\tnor1\n\t" |
| 91 | "lw\t%0, [%1]\n\t" |
| 92 | "subri\t%2, %2, 4\n\t" |
| 93 | "slli\t%2, %2, 2\n\t" |
| 94 | "lw\t%3, [%1, 4]\n\t" |
| 95 | "add\t%2, %2, %1\n\t" |
| 96 | "add\t%0, %0, %3\n\t" |
| 97 | "cmp.c\t%3, %0\n\t" |
| 98 | "lw\t%3, [%1, 8]\n\t" |
| 99 | "bleu\t1f\n\t" |
| 100 | "addi\t%0, 0x1\n\t" |
| 101 | "1:\n\t" |
| 102 | "add\t%0, %0, %3\n\t" |
| 103 | "cmp.c\t%3, %0\n\t" |
| 104 | "lw\t%3, [%1, 12]\n\t" |
| 105 | "bleu\t1f\n\t" |
| 106 | "addi\t%0, 0x1\n\t" |
| 107 | "1:add\t%0, %0, %3\n\t" |
| 108 | "cmp.c\t%3, %0\n\t" |
| 109 | "bleu\t1f\n\t" |
| 110 | "addi\t%0, 0x1\n" |
| 111 | |
| 112 | "1:\tlw\t%3, [%1, 16]\n\t" |
| 113 | "addi\t%1, 4\n\t" |
| 114 | "add\t%0, %0, %3\n\t" |
| 115 | "cmp.c\t%3, %0\n\t" |
| 116 | "bleu\t2f\n\t" |
| 117 | "addi\t%0, 0x1\n" |
| 118 | "2:cmp.c\t%2, %1\n\t" |
| 119 | "bne\t1b\n\t" |
| 120 | |
| 121 | ".set\tr1\n\t" |
| 122 | ".set optimize\n\t" |
| 123 | : "=&r" (sum), "=&r" (iph), "=&r" (ihl), "=&r" (dummy) |
| 124 | : "1" (iph), "2" (ihl)); |
| 125 | |
| 126 | return csum_fold(sum); |
| 127 | } |
| 128 | |
| 129 | static inline __wsum |
| 130 | csum_tcpudp_nofold(__be32 saddr, __be32 daddr, unsigned short len, |
| 131 | unsigned short proto, __wsum sum) |
| 132 | { |
| 133 | unsigned long tmp = (ntohs(len) << 16) + proto * 256; |
| 134 | __asm__ __volatile__( |
| 135 | ".set volatile\n\t" |
| 136 | "add\t%0, %0, %2\n\t" |
| 137 | "cmp.c\t%2, %0\n\t" |
| 138 | "bleu\t1f\n\t" |
| 139 | "addi\t%0, 0x1\n\t" |
| 140 | "1:\n\t" |
| 141 | "add\t%0, %0, %3\n\t" |
| 142 | "cmp.c\t%3, %0\n\t" |
| 143 | "bleu\t1f\n\t" |
| 144 | "addi\t%0, 0x1\n\t" |
| 145 | "1:\n\t" |
| 146 | "add\t%0, %0, %4\n\t" |
| 147 | "cmp.c\t%4, %0\n\t" |
| 148 | "bleu\t1f\n\t" |
| 149 | "addi\t%0, 0x1\n\t" |
| 150 | "1:\n\t" |
| 151 | ".set optimize\n\t" |
| 152 | : "=r" (sum) |
| 153 | : "0" (daddr), "r"(saddr), |
| 154 | "r" (tmp), |
| 155 | "r" (sum)); |
| 156 | return sum; |
| 157 | } |
| 158 | |
| 159 | /* |
| 160 | * computes the checksum of the TCP/UDP pseudo-header |
| 161 | * returns a 16-bit checksum, already complemented |
| 162 | */ |
| 163 | static inline __sum16 |
| 164 | csum_tcpudp_magic(__be32 saddr, __be32 daddr, unsigned short len, |
| 165 | unsigned short proto, __wsum sum) |
| 166 | { |
| 167 | return csum_fold(csum_tcpudp_nofold(saddr, daddr, len, proto, sum)); |
| 168 | } |
| 169 | |
| 170 | /* |
| 171 | * this routine is used for miscellaneous IP-like checksums, mainly |
| 172 | * in icmp.c |
| 173 | */ |
| 174 | |
| 175 | static inline unsigned short ip_compute_csum(const void *buff, int len) |
| 176 | { |
| 177 | return csum_fold(csum_partial(buff, len, 0)); |
| 178 | } |
| 179 | |
| 180 | #define _HAVE_ARCH_IPV6_CSUM |
| 181 | static inline __sum16 csum_ipv6_magic(const struct in6_addr *saddr, |
| 182 | const struct in6_addr *daddr, |
| 183 | __u32 len, unsigned short proto, |
| 184 | __wsum sum) |
| 185 | { |
| 186 | __asm__ __volatile__( |
| 187 | ".set\tnoreorder\t\t\t# csum_ipv6_magic\n\t" |
| 188 | ".set\tnoat\n\t" |
| 189 | "addu\t%0, %5\t\t\t# proto (long in network byte order)\n\t" |
| 190 | "sltu\t$1, %0, %5\n\t" |
| 191 | "addu\t%0, $1\n\t" |
| 192 | "addu\t%0, %6\t\t\t# csum\n\t" |
| 193 | "sltu\t$1, %0, %6\n\t" |
| 194 | "lw\t%1, 0(%2)\t\t\t# four words source address\n\t" |
| 195 | "addu\t%0, $1\n\t" |
| 196 | "addu\t%0, %1\n\t" |
| 197 | "sltu\t$1, %0, %1\n\t" |
| 198 | "lw\t%1, 4(%2)\n\t" |
| 199 | "addu\t%0, $1\n\t" |
| 200 | "addu\t%0, %1\n\t" |
| 201 | "sltu\t$1, %0, %1\n\t" |
| 202 | "lw\t%1, 8(%2)\n\t" |
| 203 | "addu\t%0, $1\n\t" |
| 204 | "addu\t%0, %1\n\t" |
| 205 | "sltu\t$1, %0, %1\n\t" |
| 206 | "lw\t%1, 12(%2)\n\t" |
| 207 | "addu\t%0, $1\n\t" |
| 208 | "addu\t%0, %1\n\t" |
| 209 | "sltu\t$1, %0, %1\n\t" |
| 210 | "lw\t%1, 0(%3)\n\t" |
| 211 | "addu\t%0, $1\n\t" |
| 212 | "addu\t%0, %1\n\t" |
| 213 | "sltu\t$1, %0, %1\n\t" |
| 214 | "lw\t%1, 4(%3)\n\t" |
| 215 | "addu\t%0, $1\n\t" |
| 216 | "addu\t%0, %1\n\t" |
| 217 | "sltu\t$1, %0, %1\n\t" |
| 218 | "lw\t%1, 8(%3)\n\t" |
| 219 | "addu\t%0, $1\n\t" |
| 220 | "addu\t%0, %1\n\t" |
| 221 | "sltu\t$1, %0, %1\n\t" |
| 222 | "lw\t%1, 12(%3)\n\t" |
| 223 | "addu\t%0, $1\n\t" |
| 224 | "addu\t%0, %1\n\t" |
| 225 | "sltu\t$1, %0, %1\n\t" |
| 226 | "addu\t%0, $1\t\t\t# Add final carry\n\t" |
| 227 | ".set\tnoat\n\t" |
| 228 | ".set\tnoreorder" |
| 229 | : "=r" (sum), "=r" (proto) |
| 230 | : "r" (saddr), "r" (daddr), |
| 231 | "0" (htonl(len)), "1" (htonl(proto)), "r" (sum)); |
| 232 | |
| 233 | return csum_fold(sum); |
| 234 | } |
| 235 | #endif /* _ASM_SCORE_CHECKSUM_H */ |