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  * accellerated 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                                    unsigned short len,
  46                                    unsigned short proto,
  47                                    __wsum sum)
  48 {
  49         return (__force __sum16)~from64to16(
  50                 (__force u64)saddr + (__force u64)daddr +
  51                 (__force u64)sum + ((len + proto) << 8));
  52 }
  53
  54 __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr,
  55                                    unsigned short len,
  56                                    unsigned short proto,
  57                                    __wsum sum)
  58 {
  59         unsigned long result;
  60
  61         result = (__force u64)saddr + (__force u64)daddr +
  62                  (__force u64)sum + ((len + proto) << 8);
  63
  64         /* Fold down to 32-bits so we don't lose in the typedef-less
  65            network stack.  */
  66         /* 64 to 33 */
  67         result = (result & 0xffffffff) + (result >> 32);
  68         /* 33 to 32 */
  69         result = (result & 0xffffffff) + (result >> 32);
  70         return (__force __wsum)result;
  71 }
  72
  73 /*
  74  * Do a 64-bit checksum on an arbitrary memory area..
  75  *
  76  * This isn't a great routine, but it's not _horrible_ either. The
  77  * inner loop could be unrolled a bit further, and there are better
  78  * ways to do the carry, but this is reasonable.
  79  */
  80 static inline unsigned long do_csum(const unsigned char * buff, int len)
  81 {
  82         int odd, count;
  83         unsigned long result = 0;
  84
  85         if (len <= 0)
  86                 goto out;
  87         odd = 1 & (unsigned long) buff;
  88         if (odd) {
  89                 result = *buff << 8;
  90                 len--;
  91                 buff++;
  92         }
  93         count = len >> 1;               /* nr of 16-bit words.. */
  94         if (count) {
  95                 if (2 & (unsigned long) buff) {
  96                         result += *(unsigned short *) buff;
  97                         count--;
  98                         len -= 2;
  99                         buff += 2;
 100                 }
 101                 count >>= 1;            /* nr of 32-bit words.. */
 102                 if (count) {
 103                         if (4 & (unsigned long) buff) {
 104                                 result += *(unsigned int *) buff;
 105                                 count--;
 106                                 len -= 4;
 107                                 buff += 4;
 108                         }
 109                         count >>= 1;    /* nr of 64-bit words.. */
 110                         if (count) {
 111                                 unsigned long carry = 0;
 112                                 do {
 113                                         unsigned long w = *(unsigned long *) buff;
 114                                         count--;
 115                                         buff += 8;
 116                                         result += carry;
 117                                         result += w;
 118                                         carry = (w > result);
 119                                 } while (count);
 120                                 result += carry;
 121                                 result = (result & 0xffffffff) + (result >> 32);
 122                         }
 123                         if (len & 4) {
 124                                 result += *(unsigned int *) buff;
 125                                 buff += 4;
 126                         }
 127                 }
 128                 if (len & 2) {
 129                         result += *(unsigned short *) buff;
 130                         buff += 2;
 131                 }
 132         }
 133         if (len & 1)
 134                 result += *buff;
 135         result = from64to16(result);
 136         if (odd)
 137                 result = ((result >> 8) & 0xff) | ((result & 0xff) << 8);
 138 out:
 139         return result;
 140 }
 141
 142 /*
 143  *      This is a version of ip_compute_csum() optimized for IP headers,
 144  *      which always checksum on 4 octet boundaries.
 145  */
 146 __sum16 ip_fast_csum(const void *iph, unsigned int ihl)
 147 {
 148         return (__force __sum16)~do_csum(iph,ihl*4);
 149 }
 150
 151 /*
 152  * computes the checksum of a memory block at buff, length len,
 153  * and adds in "sum" (32-bit)
 154  *
 155  * returns a 32-bit number suitable for feeding into itself
 156  * or csum_tcpudp_magic
 157  *
 158  * this function must be called with even lengths, except
 159  * for the last fragment, which may be odd
 160  *
 161  * it's best to have buff aligned on a 32-bit boundary
 162  */
 163 __wsum csum_partial(const void *buff, int len, __wsum sum)
 164 {
 165         unsigned long result = do_csum(buff, len);
 166
 167         /* add in old sum, and carry.. */
 168         result += (__force u32)sum;
 169         /* 32+c bits -> 32 bits */
 170         result = (result & 0xffffffff) + (result >> 32);
 171         return (__force __wsum)result;
 172 }
 173
 174 EXPORT_SYMBOL(csum_partial);
 175
 176 /*
 177  * this routine is used for miscellaneous IP-like checksums, mainly
 178  * in icmp.c
 179  */
 180 __sum16 ip_compute_csum(const void *buff, int len)
 181 {
 182         return (__force __sum16)~from64to16(do_csum(buff,len));
 183 }