Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* $Id: checksum.h,v 1.33 2002/02/01 22:01:05 davem Exp $ */ |
| 2 | #ifndef __SPARC_CHECKSUM_H |
| 3 | #define __SPARC_CHECKSUM_H |
| 4 | |
| 5 | /* checksum.h: IP/UDP/TCP checksum routines on the Sparc. |
| 6 | * |
| 7 | * Copyright(C) 1995 Linus Torvalds |
| 8 | * Copyright(C) 1995 Miguel de Icaza |
| 9 | * Copyright(C) 1996 David S. Miller |
| 10 | * Copyright(C) 1996 Eddie C. Dost |
| 11 | * Copyright(C) 1997 Jakub Jelinek |
| 12 | * |
| 13 | * derived from: |
| 14 | * Alpha checksum c-code |
| 15 | * ix86 inline assembly |
| 16 | * RFC1071 Computing the Internet Checksum |
| 17 | */ |
| 18 | |
| 19 | #include <linux/in6.h> |
| 20 | #include <asm/uaccess.h> |
| 21 | |
| 22 | /* computes the checksum of a memory block at buff, length len, |
| 23 | * and adds in "sum" (32-bit) |
| 24 | * |
| 25 | * returns a 32-bit number suitable for feeding into itself |
| 26 | * or csum_tcpudp_magic |
| 27 | * |
| 28 | * this function must be called with even lengths, except |
| 29 | * for the last fragment, which may be odd |
| 30 | * |
| 31 | * it's best to have buff aligned on a 32-bit boundary |
| 32 | */ |
| 33 | extern unsigned int csum_partial(const unsigned char * buff, int len, unsigned int sum); |
| 34 | |
| 35 | /* the same as csum_partial, but copies from fs:src while it |
| 36 | * checksums |
| 37 | * |
| 38 | * here even more important to align src and dst on a 32-bit (or even |
| 39 | * better 64-bit) boundary |
| 40 | */ |
| 41 | |
| 42 | extern unsigned int __csum_partial_copy_sparc_generic (const unsigned char *, unsigned char *); |
| 43 | |
| 44 | static inline unsigned int |
| 45 | csum_partial_copy_nocheck (const unsigned char *src, unsigned char *dst, int len, |
| 46 | unsigned int sum) |
| 47 | { |
| 48 | register unsigned int ret asm("o0") = (unsigned int)src; |
| 49 | register char *d asm("o1") = dst; |
| 50 | register int l asm("g1") = len; |
| 51 | |
| 52 | __asm__ __volatile__ ( |
| 53 | "call __csum_partial_copy_sparc_generic\n\t" |
| 54 | " mov %6, %%g7\n" |
| 55 | : "=&r" (ret), "=&r" (d), "=&r" (l) |
| 56 | : "0" (ret), "1" (d), "2" (l), "r" (sum) |
| 57 | : "o2", "o3", "o4", "o5", "o7", |
| 58 | "g2", "g3", "g4", "g5", "g7", |
| 59 | "memory", "cc"); |
| 60 | return ret; |
| 61 | } |
| 62 | |
| 63 | static inline unsigned int |
| 64 | csum_partial_copy_from_user(const unsigned char __user *src, unsigned char *dst, int len, |
| 65 | unsigned int sum, int *err) |
| 66 | { |
| 67 | if (!access_ok (VERIFY_READ, src, len)) { |
| 68 | *err = -EFAULT; |
| 69 | memset (dst, 0, len); |
| 70 | return sum; |
| 71 | } else { |
| 72 | register unsigned long ret asm("o0") = (unsigned long)src; |
| 73 | register char *d asm("o1") = dst; |
| 74 | register int l asm("g1") = len; |
| 75 | register unsigned int s asm("g7") = sum; |
| 76 | |
| 77 | __asm__ __volatile__ ( |
| 78 | ".section __ex_table,#alloc\n\t" |
| 79 | ".align 4\n\t" |
| 80 | ".word 1f,2\n\t" |
| 81 | ".previous\n" |
| 82 | "1:\n\t" |
| 83 | "call __csum_partial_copy_sparc_generic\n\t" |
| 84 | " st %8, [%%sp + 64]\n" |
| 85 | : "=&r" (ret), "=&r" (d), "=&r" (l), "=&r" (s) |
| 86 | : "0" (ret), "1" (d), "2" (l), "3" (s), "r" (err) |
| 87 | : "o2", "o3", "o4", "o5", "o7", "g2", "g3", "g4", "g5", |
| 88 | "cc", "memory"); |
| 89 | return ret; |
| 90 | } |
| 91 | } |
| 92 | |
| 93 | static inline unsigned int |
| 94 | csum_partial_copy_to_user(const unsigned char *src, unsigned char __user *dst, int len, |
| 95 | unsigned int sum, int *err) |
| 96 | { |
| 97 | if (!access_ok (VERIFY_WRITE, dst, len)) { |
| 98 | *err = -EFAULT; |
| 99 | return sum; |
| 100 | } else { |
| 101 | register unsigned long ret asm("o0") = (unsigned long)src; |
| 102 | register char __user *d asm("o1") = dst; |
| 103 | register int l asm("g1") = len; |
| 104 | register unsigned int s asm("g7") = sum; |
| 105 | |
| 106 | __asm__ __volatile__ ( |
| 107 | ".section __ex_table,#alloc\n\t" |
| 108 | ".align 4\n\t" |
| 109 | ".word 1f,1\n\t" |
| 110 | ".previous\n" |
| 111 | "1:\n\t" |
| 112 | "call __csum_partial_copy_sparc_generic\n\t" |
| 113 | " st %8, [%%sp + 64]\n" |
| 114 | : "=&r" (ret), "=&r" (d), "=&r" (l), "=&r" (s) |
| 115 | : "0" (ret), "1" (d), "2" (l), "3" (s), "r" (err) |
| 116 | : "o2", "o3", "o4", "o5", "o7", |
| 117 | "g2", "g3", "g4", "g5", |
| 118 | "cc", "memory"); |
| 119 | return ret; |
| 120 | } |
| 121 | } |
| 122 | |
| 123 | #define HAVE_CSUM_COPY_USER |
| 124 | #define csum_and_copy_to_user csum_partial_copy_to_user |
| 125 | |
| 126 | /* ihl is always 5 or greater, almost always is 5, and iph is word aligned |
| 127 | * the majority of the time. |
| 128 | */ |
| 129 | static inline unsigned short ip_fast_csum(const unsigned char *iph, |
| 130 | unsigned int ihl) |
| 131 | { |
| 132 | unsigned short sum; |
| 133 | |
| 134 | /* Note: We must read %2 before we touch %0 for the first time, |
| 135 | * because GCC can legitimately use the same register for |
| 136 | * both operands. |
| 137 | */ |
| 138 | __asm__ __volatile__("sub\t%2, 4, %%g4\n\t" |
| 139 | "ld\t[%1 + 0x00], %0\n\t" |
| 140 | "ld\t[%1 + 0x04], %%g2\n\t" |
| 141 | "ld\t[%1 + 0x08], %%g3\n\t" |
| 142 | "addcc\t%%g2, %0, %0\n\t" |
| 143 | "addxcc\t%%g3, %0, %0\n\t" |
| 144 | "ld\t[%1 + 0x0c], %%g2\n\t" |
| 145 | "ld\t[%1 + 0x10], %%g3\n\t" |
| 146 | "addxcc\t%%g2, %0, %0\n\t" |
| 147 | "addx\t%0, %%g0, %0\n" |
| 148 | "1:\taddcc\t%%g3, %0, %0\n\t" |
| 149 | "add\t%1, 4, %1\n\t" |
| 150 | "addxcc\t%0, %%g0, %0\n\t" |
| 151 | "subcc\t%%g4, 1, %%g4\n\t" |
| 152 | "be,a\t2f\n\t" |
| 153 | "sll\t%0, 16, %%g2\n\t" |
| 154 | "b\t1b\n\t" |
| 155 | "ld\t[%1 + 0x10], %%g3\n" |
| 156 | "2:\taddcc\t%0, %%g2, %%g2\n\t" |
| 157 | "srl\t%%g2, 16, %0\n\t" |
| 158 | "addx\t%0, %%g0, %0\n\t" |
| 159 | "xnor\t%%g0, %0, %0" |
| 160 | : "=r" (sum), "=&r" (iph) |
| 161 | : "r" (ihl), "1" (iph) |
| 162 | : "g2", "g3", "g4", "cc"); |
| 163 | return sum; |
| 164 | } |
| 165 | |
| 166 | /* Fold a partial checksum without adding pseudo headers. */ |
| 167 | static inline unsigned int csum_fold(unsigned int sum) |
| 168 | { |
| 169 | unsigned int tmp; |
| 170 | |
| 171 | __asm__ __volatile__("addcc\t%0, %1, %1\n\t" |
| 172 | "srl\t%1, 16, %1\n\t" |
| 173 | "addx\t%1, %%g0, %1\n\t" |
| 174 | "xnor\t%%g0, %1, %0" |
| 175 | : "=&r" (sum), "=r" (tmp) |
| 176 | : "0" (sum), "1" (sum<<16) |
| 177 | : "cc"); |
| 178 | return sum; |
| 179 | } |
| 180 | |
| 181 | static inline unsigned long csum_tcpudp_nofold(unsigned long saddr, |
| 182 | unsigned long daddr, |
| 183 | unsigned int len, |
| 184 | unsigned short proto, |
| 185 | unsigned int sum) |
| 186 | { |
| 187 | __asm__ __volatile__("addcc\t%1, %0, %0\n\t" |
| 188 | "addxcc\t%2, %0, %0\n\t" |
| 189 | "addxcc\t%3, %0, %0\n\t" |
| 190 | "addx\t%0, %%g0, %0\n\t" |
| 191 | : "=r" (sum), "=r" (saddr) |
| 192 | : "r" (daddr), "r" ((proto<<16)+len), "0" (sum), |
| 193 | "1" (saddr) |
| 194 | : "cc"); |
| 195 | return sum; |
| 196 | } |
| 197 | |
| 198 | /* |
| 199 | * computes the checksum of the TCP/UDP pseudo-header |
| 200 | * returns a 16-bit checksum, already complemented |
| 201 | */ |
| 202 | static inline unsigned short int csum_tcpudp_magic(unsigned long saddr, |
| 203 | unsigned long daddr, |
| 204 | unsigned short len, |
| 205 | unsigned short proto, |
| 206 | unsigned int sum) |
| 207 | { |
| 208 | return csum_fold(csum_tcpudp_nofold(saddr,daddr,len,proto,sum)); |
| 209 | } |
| 210 | |
| 211 | #define _HAVE_ARCH_IPV6_CSUM |
| 212 | |
| 213 | static inline unsigned short int csum_ipv6_magic(struct in6_addr *saddr, |
| 214 | struct in6_addr *daddr, |
| 215 | __u32 len, |
| 216 | unsigned short proto, |
| 217 | unsigned int sum) |
| 218 | { |
| 219 | __asm__ __volatile__ ( |
| 220 | "addcc %3, %4, %%g4\n\t" |
| 221 | "addxcc %5, %%g4, %%g4\n\t" |
| 222 | "ld [%2 + 0x0c], %%g2\n\t" |
| 223 | "ld [%2 + 0x08], %%g3\n\t" |
| 224 | "addxcc %%g2, %%g4, %%g4\n\t" |
| 225 | "ld [%2 + 0x04], %%g2\n\t" |
| 226 | "addxcc %%g3, %%g4, %%g4\n\t" |
| 227 | "ld [%2 + 0x00], %%g3\n\t" |
| 228 | "addxcc %%g2, %%g4, %%g4\n\t" |
| 229 | "ld [%1 + 0x0c], %%g2\n\t" |
| 230 | "addxcc %%g3, %%g4, %%g4\n\t" |
| 231 | "ld [%1 + 0x08], %%g3\n\t" |
| 232 | "addxcc %%g2, %%g4, %%g4\n\t" |
| 233 | "ld [%1 + 0x04], %%g2\n\t" |
| 234 | "addxcc %%g3, %%g4, %%g4\n\t" |
| 235 | "ld [%1 + 0x00], %%g3\n\t" |
| 236 | "addxcc %%g2, %%g4, %%g4\n\t" |
| 237 | "addxcc %%g3, %%g4, %0\n\t" |
| 238 | "addx 0, %0, %0\n" |
| 239 | : "=&r" (sum) |
| 240 | : "r" (saddr), "r" (daddr), |
| 241 | "r"(htonl(len)), "r"(htonl(proto)), "r"(sum) |
| 242 | : "g2", "g3", "g4", "cc"); |
| 243 | |
| 244 | return csum_fold(sum); |
| 245 | } |
| 246 | |
| 247 | /* this routine is used for miscellaneous IP-like checksums, mainly in icmp.c */ |
| 248 | static inline unsigned short ip_compute_csum(unsigned char * buff, int len) |
| 249 | { |
| 250 | return csum_fold(csum_partial(buff, len, 0)); |
| 251 | } |
| 252 | |
| 253 | #endif /* !(__SPARC_CHECKSUM_H) */ |