[9b9c4dfd] | 1 | /* |
---|
| 2 | * Checksum routine for Internet Protocol family headers. |
---|
| 3 | * |
---|
| 4 | * This routine is very heavily used in the network |
---|
| 5 | * code and should be modified for each CPU to be as fast as possible. |
---|
| 6 | * |
---|
| 7 | * This implementation is the PowerPC version. |
---|
| 8 | * |
---|
| 9 | * $Id$ |
---|
| 10 | */ |
---|
| 11 | |
---|
| 12 | #include <stdio.h> /* for puts */ |
---|
| 13 | |
---|
| 14 | #undef ADDCARRY |
---|
| 15 | #define ADDCARRY(x) if ((x) > 0xffff) (x) -= 0xffff |
---|
| 16 | #define REDUCE {sum = (sum & 0xffff) + (sum >> 16); ADDCARRY(sum);} |
---|
| 17 | |
---|
| 18 | /* |
---|
| 19 | * Thanks to gcc we don't have to guess |
---|
| 20 | * which registers contain sum & w. |
---|
| 21 | */ |
---|
| 22 | |
---|
| 23 | #define LDTMP(n) tmp = *((u_int *)((u_char *)w + n)) |
---|
| 24 | |
---|
| 25 | #define ADD(n) \ |
---|
| 26 | LDTMP(n); \ |
---|
| 27 | __asm__ volatile("addc %0,%0,%2" : "=r" (sum) : "0" (sum), "r" (tmp)) |
---|
| 28 | |
---|
| 29 | #define ADDC(n) \ |
---|
| 30 | LDTMP(n); \ |
---|
| 31 | __asm__ volatile("adde %0,%0,%2" : "=r" (sum) : "0" (sum), "r" (tmp)) |
---|
| 32 | |
---|
| 33 | #define MOP \ |
---|
| 34 | tmp = 0; \ |
---|
| 35 | __asm__ volatile("adde %0,%0,%2" : "=r" (sum) : "0" (sum), "r" (tmp)) |
---|
| 36 | |
---|
| 37 | #define LOAD(n) junk = (u_char) *((volatile u_char *) w + n) |
---|
| 38 | |
---|
| 39 | |
---|
| 40 | int |
---|
| 41 | in_cksum(m, len) |
---|
| 42 | register struct mbuf *m; |
---|
| 43 | register int len; |
---|
| 44 | { |
---|
| 45 | register u_short *w; |
---|
| 46 | register unsigned sum = 0; |
---|
| 47 | register unsigned tmp; |
---|
| 48 | register int mlen = 0; |
---|
| 49 | int byte_swapped = 0; |
---|
| 50 | union { char c[2]; u_short s; } su; |
---|
| 51 | |
---|
| 52 | for (;m && len; m = m->m_next) { |
---|
| 53 | if (m->m_len == 0) |
---|
| 54 | continue; |
---|
| 55 | w = mtod(m, u_short *); |
---|
| 56 | if (mlen == -1) { |
---|
| 57 | /* |
---|
| 58 | * The first byte of this mbuf is the continuation |
---|
| 59 | * of a word spanning between this mbuf and the |
---|
| 60 | * last mbuf. |
---|
| 61 | */ |
---|
| 62 | |
---|
| 63 | /* su.c[0] is already saved when scanning previous |
---|
| 64 | * mbuf. sum was REDUCEd when we found mlen == -1 |
---|
| 65 | */ |
---|
| 66 | su.c[1] = *(u_char *)w; |
---|
| 67 | sum += su.s; |
---|
| 68 | w = (u_short *)((char *)w + 1); |
---|
| 69 | mlen = m->m_len - 1; |
---|
| 70 | len--; |
---|
| 71 | } else |
---|
| 72 | mlen = m->m_len; |
---|
| 73 | if (len < mlen) |
---|
| 74 | mlen = len; |
---|
| 75 | len -= mlen; |
---|
| 76 | /* |
---|
| 77 | * Force to long boundary so we do longword aligned |
---|
| 78 | * memory operations |
---|
| 79 | */ |
---|
| 80 | if (3 & (int) w) { |
---|
| 81 | REDUCE; |
---|
| 82 | if ((1 & (int) w) && (mlen > 0)) { |
---|
| 83 | sum <<= 8; |
---|
| 84 | su.c[0] = *(char *)w; |
---|
| 85 | w = (u_short *)((char *)w + 1); |
---|
| 86 | mlen--; |
---|
| 87 | byte_swapped = 1; |
---|
| 88 | } |
---|
| 89 | if ((2 & (int) w) && (mlen >= 2)) { |
---|
| 90 | sum += *w++; |
---|
| 91 | mlen -= 2; |
---|
| 92 | } |
---|
| 93 | } |
---|
| 94 | /* |
---|
| 95 | * Advance to a cache line boundary. |
---|
| 96 | */ |
---|
| 97 | if (4 & (int) w && mlen >= 4) { |
---|
| 98 | ADD(0); |
---|
| 99 | MOP; |
---|
| 100 | w += 2; |
---|
| 101 | mlen -= 4; |
---|
| 102 | } |
---|
| 103 | if (8 & (int) w && mlen >= 8) { |
---|
| 104 | ADD(0); |
---|
| 105 | ADDC(4); |
---|
| 106 | MOP; |
---|
| 107 | w += 4; |
---|
| 108 | mlen -= 8; |
---|
| 109 | } |
---|
| 110 | /* |
---|
| 111 | * Do as much of the checksum as possible 32 bits at at time. |
---|
| 112 | * In fact, this loop is unrolled to make overhead from |
---|
| 113 | * branches &c small. |
---|
| 114 | */ |
---|
| 115 | mlen -= 1; |
---|
| 116 | while ((mlen -= 32) >= 0) { |
---|
| 117 | u_char junk; |
---|
| 118 | /* |
---|
| 119 | * Add with carry 16 words and fold in the last |
---|
| 120 | * carry by adding a 0 with carry. |
---|
| 121 | * |
---|
| 122 | * The early ADD(16) and the LOAD(32) are intended |
---|
| 123 | * to help get the data into the cache. |
---|
| 124 | */ |
---|
| 125 | ADD(16); |
---|
| 126 | ADDC(0); |
---|
| 127 | ADDC(4); |
---|
| 128 | ADDC(8); |
---|
| 129 | ADDC(12); |
---|
| 130 | LOAD(32); |
---|
| 131 | ADDC(20); |
---|
| 132 | ADDC(24); |
---|
| 133 | ADDC(28); |
---|
| 134 | MOP; |
---|
| 135 | w += 16; |
---|
| 136 | } |
---|
| 137 | mlen += 32 + 1; |
---|
| 138 | if (mlen >= 32) { |
---|
| 139 | ADD(16); |
---|
| 140 | ADDC(0); |
---|
| 141 | ADDC(4); |
---|
| 142 | ADDC(8); |
---|
| 143 | ADDC(12); |
---|
| 144 | ADDC(20); |
---|
| 145 | ADDC(24); |
---|
| 146 | ADDC(28); |
---|
| 147 | MOP; |
---|
| 148 | w += 16; |
---|
| 149 | mlen -= 32; |
---|
| 150 | } |
---|
| 151 | if (mlen >= 16) { |
---|
| 152 | ADD(0); |
---|
| 153 | ADDC(4); |
---|
| 154 | ADDC(8); |
---|
| 155 | ADDC(12); |
---|
| 156 | MOP; |
---|
| 157 | w += 8; |
---|
| 158 | mlen -= 16; |
---|
| 159 | } |
---|
| 160 | if (mlen >= 8) { |
---|
| 161 | ADD(0); |
---|
| 162 | ADDC(4); |
---|
| 163 | MOP; |
---|
| 164 | w += 4; |
---|
| 165 | mlen -= 8; |
---|
| 166 | } |
---|
| 167 | if (mlen == 0 && byte_swapped == 0) |
---|
| 168 | continue; /* worth 1% maybe ?? */ |
---|
| 169 | REDUCE; |
---|
| 170 | while ((mlen -= 2) >= 0) { |
---|
| 171 | sum += *w++; |
---|
| 172 | } |
---|
| 173 | if (byte_swapped) { |
---|
| 174 | sum <<= 8; |
---|
| 175 | byte_swapped = 0; |
---|
| 176 | if (mlen == -1) { |
---|
| 177 | su.c[1] = *(char *)w; |
---|
| 178 | sum += su.s; |
---|
| 179 | mlen = 0; |
---|
| 180 | } else |
---|
| 181 | mlen = -1; |
---|
| 182 | } else if (mlen == -1) |
---|
| 183 | /* |
---|
| 184 | * This mbuf has odd number of bytes. |
---|
| 185 | * There could be a word split betwen |
---|
| 186 | * this mbuf and the next mbuf. |
---|
| 187 | * Save the last byte (to prepend to next mbuf). |
---|
| 188 | */ |
---|
| 189 | su.c[0] = *(char *)w; |
---|
| 190 | } |
---|
| 191 | |
---|
| 192 | if (len) |
---|
| 193 | puts("cksum: out of data"); |
---|
| 194 | if (mlen == -1) { |
---|
| 195 | /* The last mbuf has odd # of bytes. Follow the |
---|
| 196 | standard (the odd byte is shifted left by 8 bits) */ |
---|
| 197 | su.c[1] = 0; |
---|
| 198 | sum += su.s; |
---|
| 199 | } |
---|
| 200 | REDUCE; |
---|
| 201 | return (~sum & 0xffff); |
---|
| 202 | } |
---|