]>
Commit | Line | Data |
---|---|---|
1 | // SPDX-License-Identifier: GPL-2.0 | |
2 | /* | |
3 | * arch/alpha/lib/checksum.c | |
4 | * | |
5 | * This file contains network checksum routines that are better done | |
6 | * in an architecture-specific manner due to speed.. | |
7 | * Comments in other versions indicate that the algorithms are from RFC1071 | |
8 | * | |
9 | * accelerated versions (and 21264 assembly versions ) contributed by | |
10 | * Rick Gorton <rick.gorton@alpha-processor.com> | |
11 | */ | |
12 | ||
13 | #include <linux/module.h> | |
14 | #include <linux/string.h> | |
15 | ||
16 | #include <asm/byteorder.h> | |
17 | ||
18 | static inline unsigned short from64to16(unsigned long x) | |
19 | { | |
20 | /* Using extract instructions is a bit more efficient | |
21 | than the original shift/bitmask version. */ | |
22 | ||
23 | union { | |
24 | unsigned long ul; | |
25 | unsigned int ui[2]; | |
26 | unsigned short us[4]; | |
27 | } in_v, tmp_v, out_v; | |
28 | ||
29 | in_v.ul = x; | |
30 | tmp_v.ul = (unsigned long) in_v.ui[0] + (unsigned long) in_v.ui[1]; | |
31 | ||
32 | /* Since the bits of tmp_v.sh[3] are going to always be zero, | |
33 | we don't have to bother to add that in. */ | |
34 | out_v.ul = (unsigned long) tmp_v.us[0] + (unsigned long) tmp_v.us[1] | |
35 | + (unsigned long) tmp_v.us[2]; | |
36 | ||
37 | /* Similarly, out_v.us[2] is always zero for the final add. */ | |
38 | return out_v.us[0] + out_v.us[1]; | |
39 | } | |
40 | ||
41 | /* | |
42 | * computes the checksum of the TCP/UDP pseudo-header | |
43 | * returns a 16-bit checksum, already complemented. | |
44 | */ | |
45 | __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr, | |
46 | __u32 len, __u8 proto, __wsum sum) | |
47 | { | |
48 | return (__force __sum16)~from64to16( | |
49 | (__force u64)saddr + (__force u64)daddr + | |
50 | (__force u64)sum + ((len + proto) << 8)); | |
51 | } | |
52 | EXPORT_SYMBOL(csum_tcpudp_magic); | |
53 | ||
54 | __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr, | |
55 | __u32 len, __u8 proto, __wsum sum) | |
56 | { | |
57 | unsigned long result; | |
58 | ||
59 | result = (__force u64)saddr + (__force u64)daddr + | |
60 | (__force u64)sum + ((len + proto) << 8); | |
61 | ||
62 | /* Fold down to 32-bits so we don't lose in the typedef-less | |
63 | network stack. */ | |
64 | /* 64 to 33 */ | |
65 | result = (result & 0xffffffff) + (result >> 32); | |
66 | /* 33 to 32 */ | |
67 | result = (result & 0xffffffff) + (result >> 32); | |
68 | return (__force __wsum)result; | |
69 | } | |
70 | EXPORT_SYMBOL(csum_tcpudp_nofold); | |
71 | ||
72 | /* | |
73 | * Do a 64-bit checksum on an arbitrary memory area.. | |
74 | * | |
75 | * This isn't a great routine, but it's not _horrible_ either. The | |
76 | * inner loop could be unrolled a bit further, and there are better | |
77 | * ways to do the carry, but this is reasonable. | |
78 | */ | |
79 | static inline unsigned long do_csum(const unsigned char * buff, int len) | |
80 | { | |
81 | int odd, count; | |
82 | unsigned long result = 0; | |
83 | ||
84 | if (len <= 0) | |
85 | goto out; | |
86 | odd = 1 & (unsigned long) buff; | |
87 | if (odd) { | |
88 | result = *buff << 8; | |
89 | len--; | |
90 | buff++; | |
91 | } | |
92 | count = len >> 1; /* nr of 16-bit words.. */ | |
93 | if (count) { | |
94 | if (2 & (unsigned long) buff) { | |
95 | result += *(unsigned short *) buff; | |
96 | count--; | |
97 | len -= 2; | |
98 | buff += 2; | |
99 | } | |
100 | count >>= 1; /* nr of 32-bit words.. */ | |
101 | if (count) { | |
102 | if (4 & (unsigned long) buff) { | |
103 | result += *(unsigned int *) buff; | |
104 | count--; | |
105 | len -= 4; | |
106 | buff += 4; | |
107 | } | |
108 | count >>= 1; /* nr of 64-bit words.. */ | |
109 | if (count) { | |
110 | unsigned long carry = 0; | |
111 | do { | |
112 | unsigned long w = *(unsigned long *) buff; | |
113 | count--; | |
114 | buff += 8; | |
115 | result += carry; | |
116 | result += w; | |
117 | carry = (w > result); | |
118 | } while (count); | |
119 | result += carry; | |
120 | result = (result & 0xffffffff) + (result >> 32); | |
121 | } | |
122 | if (len & 4) { | |
123 | result += *(unsigned int *) buff; | |
124 | buff += 4; | |
125 | } | |
126 | } | |
127 | if (len & 2) { | |
128 | result += *(unsigned short *) buff; | |
129 | buff += 2; | |
130 | } | |
131 | } | |
132 | if (len & 1) | |
133 | result += *buff; | |
134 | result = from64to16(result); | |
135 | if (odd) | |
136 | result = ((result >> 8) & 0xff) | ((result & 0xff) << 8); | |
137 | out: | |
138 | return result; | |
139 | } | |
140 | ||
141 | /* | |
142 | * This is a version of ip_compute_csum() optimized for IP headers, | |
143 | * which always checksum on 4 octet boundaries. | |
144 | */ | |
145 | __sum16 ip_fast_csum(const void *iph, unsigned int ihl) | |
146 | { | |
147 | return (__force __sum16)~do_csum(iph,ihl*4); | |
148 | } | |
149 | EXPORT_SYMBOL(ip_fast_csum); | |
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 | } | |
184 | EXPORT_SYMBOL(ip_compute_csum); |