arch/alpha/lib/checksum.c

   1 /*
   2  * arch/alpha/lib/checksum.c
   3  *
   4  * This file contains network checksum routines that are better done
   5  * in an architecture-specific manner due to speed..
   6  * Comments in other versions indicate that the algorithms are from RFC1071
   7  *
   8  * accelerated versions (and 21264 assembly versions ) contributed by
   9  *      Rick Gorton     <rick.gorton@alpha-processor.com>
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/string.h>
  14
  15 #include <asm/byteorder.h>
  16
  17 static inline unsigned short from64to16(unsigned long x)
  18 {
  19         /* Using extract instructions is a bit more efficient
  20            than the original shift/bitmask version.  */
  21
  22         union {
  23                 unsigned long   ul;
  24                 unsigned int    ui[2];
  25                 unsigned short  us[4];
  26         } in_v, tmp_v, out_v;
  27
  28         in_v.ul = x;
  29         tmp_v.ul = (unsigned long) in_v.ui[0] + (unsigned long) in_v.ui[1];
  30
  31         /* Since the bits of tmp_v.sh[3] are going to always be zero,
  32            we don't have to bother to add that in.  */
  33         out_v.ul = (unsigned long) tmp_v.us[0] + (unsigned long) tmp_v.us[1]
  34                         + (unsigned long) tmp_v.us[2];
  35
  36         /* Similarly, out_v.us[2] is always zero for the final add.  */
  37         return out_v.us[0] + out_v.us[1];
  38 }
  39
  40 /*
  41  * computes the checksum of the TCP/UDP pseudo-header
  42  * returns a 16-bit checksum, already complemented.
  43  */
  44 __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr,
  45                           __u32 len, __u8 proto, __wsum sum)
  46 {
  47         return (__force __sum16)~from64to16(
  48                 (__force u64)saddr + (__force u64)daddr +
  49                 (__force u64)sum + ((len + proto) << 8));
  50 }
  51
  52 __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr,
  53                           __u32 len, __u8 proto, __wsum sum)
  54 {
  55         unsigned long result;
  56
  57         result = (__force u64)saddr + (__force u64)daddr +
  58                  (__force u64)sum + ((len + proto) << 8);
  59
  60         /* Fold down to 32-bits so we don't lose in the typedef-less
  61            network stack.  */
  62         /* 64 to 33 */
  63         result = (result & 0xffffffff) + (result >> 32);
  64         /* 33 to 32 */
  65         result = (result & 0xffffffff) + (result >> 32);
  66         return (__force __wsum)result;
  67 }
  68 EXPORT_SYMBOL(csum_tcpudp_nofold);
  69
  70 /*
  71  * Do a 64-bit checksum on an arbitrary memory area..
  72  *
  73  * This isn't a great routine, but it's not _horrible_ either. The
  74  * inner loop could be unrolled a bit further, and there are better
  75  * ways to do the carry, but this is reasonable.
  76  */
  77 static inline unsigned long do_csum(const unsigned char * buff, int len)
  78 {
  79         int odd, count;
  80         unsigned long result = 0;
  81
  82         if (len <= 0)
  83                 goto out;
  84         odd = 1 & (unsigned long) buff;
  85         if (odd) {
  86                 result = *buff << 8;
  87                 len--;
  88                 buff++;
  89         }
  90         count = len >> 1;               /* nr of 16-bit words.. */
  91         if (count) {
  92                 if (2 & (unsigned long) buff) {
  93                         result += *(unsigned short *) buff;
  94                         count--;
  95                         len -= 2;
  96                         buff += 2;
  97                 }
  98                 count >>= 1;            /* nr of 32-bit words.. */
  99                 if (count) {
 100                         if (4 & (unsigned long) buff) {
 101                                 result += *(unsigned int *) buff;
 102                                 count--;
 103                                 len -= 4;
 104                                 buff += 4;
 105                         }
 106                         count >>= 1;    /* nr of 64-bit words.. */
 107                         if (count) {
 108                                 unsigned long carry = 0;
 109                                 do {
 110                                         unsigned long w = *(unsigned long *) buff;
 111                                         count--;
 112                                         buff += 8;
 113                                         result += carry;
 114                                         result += w;
 115                                         carry = (w > result);
 116                                 } while (count);
 117                                 result += carry;
 118                                 result = (result & 0xffffffff) + (result >> 32);
 119                         }
 120                         if (len & 4) {
 121                                 result += *(unsigned int *) buff;
 122                                 buff += 4;
 123                         }
 124                 }
 125                 if (len & 2) {
 126                         result += *(unsigned short *) buff;
 127                         buff += 2;
 128                 }
 129         }
 130         if (len & 1)
 131                 result += *buff;
 132         result = from64to16(result);
 133         if (odd)
 134                 result = ((result >> 8) & 0xff) | ((result & 0xff) << 8);
 135 out:
 136         return result;
 137 }
 138
 139 /*
 140  *      This is a version of ip_compute_csum() optimized for IP headers,
 141  *      which always checksum on 4 octet boundaries.
 142  */
 143 __sum16 ip_fast_csum(const void *iph, unsigned int ihl)
 144 {
 145         return (__force __sum16)~do_csum(iph,ihl*4);
 146 }
 147
 148 /*
 149  * computes the checksum of a memory block at buff, length len,
 150  * and adds in "sum" (32-bit)
 151  *
 152  * returns a 32-bit number suitable for feeding into itself
 153  * or csum_tcpudp_magic
 154  *
 155  * this function must be called with even lengths, except
 156  * for the last fragment, which may be odd
 157  *
 158  * it's best to have buff aligned on a 32-bit boundary
 159  */
 160 __wsum csum_partial(const void *buff, int len, __wsum sum)
 161 {
 162         unsigned long result = do_csum(buff, len);
 163
 164         /* add in old sum, and carry.. */
 165         result += (__force u32)sum;
 166         /* 32+c bits -> 32 bits */
 167         result = (result & 0xffffffff) + (result >> 32);
 168         return (__force __wsum)result;
 169 }
 170
 171 EXPORT_SYMBOL(csum_partial);
 172
 173 /*
 174  * this routine is used for miscellaneous IP-like checksums, mainly
 175  * in icmp.c
 176  */
 177 __sum16 ip_compute_csum(const void *buff, int len)
 178 {
 179         return (__force __sum16)~from64to16(do_csum(buff,len));
 180 }