arch/sparc/include/asm/checksum_32.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   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 <linux/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 __wsum csum_partial(const void *buff, int len, __wsum 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 unsigned int __csum_partial_copy_sparc_generic (const unsigned char *, unsigned char *);
  43
  44 static inline __wsum
  45 csum_partial_copy_nocheck(const void *src, void *dst, int len)
  46 {
  47         register unsigned int ret asm("o0") = (unsigned int)src;
  48         register char *d asm("o1") = dst;
  49         register int l asm("g1") = len;
  50
  51         __asm__ __volatile__ (
  52                 "call __csum_partial_copy_sparc_generic\n\t"
  53                 " mov -1, %%g7\n"
  54         : "=&r" (ret), "=&r" (d), "=&r" (l)
  55         : "0" (ret), "1" (d), "2" (l)
  56         : "o2", "o3", "o4", "o5", "o7",
  57           "g2", "g3", "g4", "g5", "g7",
  58           "memory", "cc");
  59         return (__force __wsum)ret;
  60 }
  61
  62 static inline __wsum
  63 csum_and_copy_from_user(const void __user *src, void *dst, int len)
  64 {
  65         if (unlikely(!access_ok(src, len)))
  66                 return 0;
  67         return csum_partial_copy_nocheck((__force void *)src, dst, len);
  68 }
  69
  70 static inline __wsum
  71 csum_and_copy_to_user(const void *src, void __user *dst, int len)
  72 {
  73         if (!access_ok(dst, len))
  74                 return 0;
  75         return csum_partial_copy_nocheck(src, (__force void *)dst, len);
  76 }
  77
  78 /* ihl is always 5 or greater, almost always is 5, and iph is word aligned
  79  * the majority of the time.
  80  */
  81 static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl)
  82 {
  83         __sum16 sum;
  84
  85         /* Note: We must read %2 before we touch %0 for the first time,
  86          *       because GCC can legitimately use the same register for
  87          *       both operands.
  88          */
  89         __asm__ __volatile__("sub\t%2, 4, %%g4\n\t"
  90                              "ld\t[%1 + 0x00], %0\n\t"
  91                              "ld\t[%1 + 0x04], %%g2\n\t"
  92                              "ld\t[%1 + 0x08], %%g3\n\t"
  93                              "addcc\t%%g2, %0, %0\n\t"
  94                              "addxcc\t%%g3, %0, %0\n\t"
  95                              "ld\t[%1 + 0x0c], %%g2\n\t"
  96                              "ld\t[%1 + 0x10], %%g3\n\t"
  97                              "addxcc\t%%g2, %0, %0\n\t"
  98                              "addx\t%0, %%g0, %0\n"
  99                              "1:\taddcc\t%%g3, %0, %0\n\t"
 100                              "add\t%1, 4, %1\n\t"
 101                              "addxcc\t%0, %%g0, %0\n\t"
 102                              "subcc\t%%g4, 1, %%g4\n\t"
 103                              "be,a\t2f\n\t"
 104                              "sll\t%0, 16, %%g2\n\t"
 105                              "b\t1b\n\t"
 106                              "ld\t[%1 + 0x10], %%g3\n"
 107                              "2:\taddcc\t%0, %%g2, %%g2\n\t"
 108                              "srl\t%%g2, 16, %0\n\t"
 109                              "addx\t%0, %%g0, %0\n\t"
 110                              "xnor\t%%g0, %0, %0"
 111                              : "=r" (sum), "=&r" (iph)
 112                              : "r" (ihl), "1" (iph)
 113                              : "g2", "g3", "g4", "cc", "memory");
 114         return sum;
 115 }
 116
 117 /* Fold a partial checksum without adding pseudo headers. */
 118 static inline __sum16 csum_fold(__wsum sum)
 119 {
 120         unsigned int tmp;
 121
 122         __asm__ __volatile__("addcc\t%0, %1, %1\n\t"
 123                              "srl\t%1, 16, %1\n\t"
 124                              "addx\t%1, %%g0, %1\n\t"
 125                              "xnor\t%%g0, %1, %0"
 126                              : "=&r" (sum), "=r" (tmp)
 127                              : "0" (sum), "1" ((__force u32)sum<<16)
 128                              : "cc");
 129         return (__force __sum16)sum;
 130 }
 131
 132 static inline __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr,
 133                                         __u32 len, __u8 proto,
 134                                         __wsum sum)
 135 {
 136         __asm__ __volatile__("addcc\t%1, %0, %0\n\t"
 137                              "addxcc\t%2, %0, %0\n\t"
 138                              "addxcc\t%3, %0, %0\n\t"
 139                              "addx\t%0, %%g0, %0\n\t"
 140                              : "=r" (sum), "=r" (saddr)
 141                              : "r" (daddr), "r" (proto + len), "0" (sum),
 142                                "1" (saddr)
 143                              : "cc");
 144         return sum;
 145 }
 146
 147 /*
 148  * computes the checksum of the TCP/UDP pseudo-header
 149  * returns a 16-bit checksum, already complemented
 150  */
 151 static inline __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr,
 152                                         __u32 len, __u8 proto,
 153                                         __wsum sum)
 154 {
 155         return csum_fold(csum_tcpudp_nofold(saddr,daddr,len,proto,sum));
 156 }
 157
 158 #define _HAVE_ARCH_IPV6_CSUM
 159
 160 static inline __sum16 csum_ipv6_magic(const struct in6_addr *saddr,
 161                                       const struct in6_addr *daddr,
 162                                       __u32 len, __u8 proto, __wsum sum)
 163 {
 164         __asm__ __volatile__ (
 165                 "addcc  %3, %4, %%g4\n\t"
 166                 "addxcc %5, %%g4, %%g4\n\t"
 167                 "ld     [%2 + 0x0c], %%g2\n\t"
 168                 "ld     [%2 + 0x08], %%g3\n\t"
 169                 "addxcc %%g2, %%g4, %%g4\n\t"
 170                 "ld     [%2 + 0x04], %%g2\n\t"
 171                 "addxcc %%g3, %%g4, %%g4\n\t"
 172                 "ld     [%2 + 0x00], %%g3\n\t"
 173                 "addxcc %%g2, %%g4, %%g4\n\t"
 174                 "ld     [%1 + 0x0c], %%g2\n\t"
 175                 "addxcc %%g3, %%g4, %%g4\n\t"
 176                 "ld     [%1 + 0x08], %%g3\n\t"
 177                 "addxcc %%g2, %%g4, %%g4\n\t"
 178                 "ld     [%1 + 0x04], %%g2\n\t"
 179                 "addxcc %%g3, %%g4, %%g4\n\t"
 180                 "ld     [%1 + 0x00], %%g3\n\t"
 181                 "addxcc %%g2, %%g4, %%g4\n\t"
 182                 "addxcc %%g3, %%g4, %0\n\t"
 183                 "addx   0, %0, %0\n"
 184                 : "=&r" (sum)
 185                 : "r" (saddr), "r" (daddr),
 186                   "r"(htonl(len)), "r"(htonl(proto)), "r"(sum)
 187                 : "g2", "g3", "g4", "cc");
 188
 189         return csum_fold(sum);
 190 }
 191
 192 /* this routine is used for miscellaneous IP-like checksums, mainly in icmp.c */
 193 static inline __sum16 ip_compute_csum(const void *buff, int len)
 194 {
 195         return csum_fold(csum_partial(buff, len, 0));
 196 }
 197
 198 #define HAVE_ARCH_CSUM_ADD
 199 static inline __wsum csum_add(__wsum csum, __wsum addend)
 200 {
 201         __asm__ __volatile__(
 202                 "addcc   %0, %1, %0\n"
 203                 "addx    %0, %%g0, %0"
 204                 : "=r" (csum)
 205                 : "r" (addend), "0" (csum));
 206
 207         return csum;
 208 }
 209
 210 #endif /* !(__SPARC_CHECKSUM_H) */