2 * arch/alpha/lib/checksum.c
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
8 * accelerated versions (and 21264 assembly versions ) contributed by
9 * Rick Gorton <rick.gorton@alpha-processor.com>
12 #include <linux/module.h>
13 #include <linux/string.h>
15 #include <asm/byteorder.h>
17 static inline unsigned short from64to16(unsigned long x)
19 /* Using extract instructions is a bit more efficient
20 than the original shift/bitmask version. */
29 tmp_v.ul = (unsigned long) in_v.ui[0] + (unsigned long) in_v.ui[1];
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];
36 /* Similarly, out_v.us[2] is always zero for the final add. */
37 return out_v.us[0] + out_v.us[1];
41 * computes the checksum of the TCP/UDP pseudo-header
42 * returns a 16-bit checksum, already complemented.
44 __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr,
45 __u32 len, __u8 proto, __wsum sum)
47 return (__force __sum16)~from64to16(
48 (__force u64)saddr + (__force u64)daddr +
49 (__force u64)sum + ((len + proto) << 8));
51 EXPORT_SYMBOL(csum_tcpudp_magic);
53 __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr,
54 __u32 len, __u8 proto, __wsum sum)
58 result = (__force u64)saddr + (__force u64)daddr +
59 (__force u64)sum + ((len + proto) << 8);
61 /* Fold down to 32-bits so we don't lose in the typedef-less
64 result = (result & 0xffffffff) + (result >> 32);
66 result = (result & 0xffffffff) + (result >> 32);
67 return (__force __wsum)result;
69 EXPORT_SYMBOL(csum_tcpudp_nofold);
72 * Do a 64-bit checksum on an arbitrary memory area..
74 * This isn't a great routine, but it's not _horrible_ either. The
75 * inner loop could be unrolled a bit further, and there are better
76 * ways to do the carry, but this is reasonable.
78 static inline unsigned long do_csum(const unsigned char * buff, int len)
81 unsigned long result = 0;
85 odd = 1 & (unsigned long) buff;
91 count = len >> 1; /* nr of 16-bit words.. */
93 if (2 & (unsigned long) buff) {
94 result += *(unsigned short *) buff;
99 count >>= 1; /* nr of 32-bit words.. */
101 if (4 & (unsigned long) buff) {
102 result += *(unsigned int *) buff;
107 count >>= 1; /* nr of 64-bit words.. */
109 unsigned long carry = 0;
111 unsigned long w = *(unsigned long *) buff;
116 carry = (w > result);
119 result = (result & 0xffffffff) + (result >> 32);
122 result += *(unsigned int *) buff;
127 result += *(unsigned short *) buff;
133 result = from64to16(result);
135 result = ((result >> 8) & 0xff) | ((result & 0xff) << 8);
141 * This is a version of ip_compute_csum() optimized for IP headers,
142 * which always checksum on 4 octet boundaries.
144 __sum16 ip_fast_csum(const void *iph, unsigned int ihl)
146 return (__force __sum16)~do_csum(iph,ihl*4);
148 EXPORT_SYMBOL(ip_fast_csum);
151 * computes the checksum of a memory block at buff, length len,
152 * and adds in "sum" (32-bit)
154 * returns a 32-bit number suitable for feeding into itself
155 * or csum_tcpudp_magic
157 * this function must be called with even lengths, except
158 * for the last fragment, which may be odd
160 * it's best to have buff aligned on a 32-bit boundary
162 __wsum csum_partial(const void *buff, int len, __wsum sum)
164 unsigned long result = do_csum(buff, len);
166 /* add in old sum, and carry.. */
167 result += (__force u32)sum;
168 /* 32+c bits -> 32 bits */
169 result = (result & 0xffffffff) + (result >> 32);
170 return (__force __wsum)result;
173 EXPORT_SYMBOL(csum_partial);
176 * this routine is used for miscellaneous IP-like checksums, mainly
179 __sum16 ip_compute_csum(const void *buff, int len)
181 return (__force __sum16)~from64to16(do_csum(buff,len));
183 EXPORT_SYMBOL(ip_compute_csum);