| 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) */ |