]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
f5e706ad SR |
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> | |
7c0f6ba6 | 20 | #include <linux/uaccess.h> |
f5e706ad SR |
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 | */ | |
f05a6865 | 33 | __wsum csum_partial(const void *buff, int len, __wsum sum); |
f5e706ad SR |
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 | ||
f05a6865 | 42 | unsigned int __csum_partial_copy_sparc_generic (const unsigned char *, unsigned char *); |
f5e706ad SR |
43 | |
44 | static inline __wsum | |
cc44c17b | 45 | csum_partial_copy_nocheck(const void *src, void *dst, int len) |
f5e706ad SR |
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 %6, %%g7\n" | |
54 | : "=&r" (ret), "=&r" (d), "=&r" (l) | |
cc44c17b | 55 | : "0" (ret), "1" (d), "2" (l), "r" (0) |
f5e706ad SR |
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 | |
76666be8 | 63 | csum_and_copy_from_user(const void __user *src, void *dst, int len, |
f5e706ad SR |
64 | __wsum sum, int *err) |
65 | { | |
66 | register unsigned long ret asm("o0") = (unsigned long)src; | |
67 | register char *d asm("o1") = dst; | |
68 | register int l asm("g1") = len; | |
69 | register __wsum s asm("g7") = sum; | |
70 | ||
76666be8 AV |
71 | if (unlikely(!access_ok(src, len))) { |
72 | if (len) | |
73 | *err = -EFAULT; | |
74 | return sum; | |
75 | } | |
76 | ||
f5e706ad SR |
77 | __asm__ __volatile__ ( |
78 | ".section __ex_table,#alloc\n\t" | |
79 | ".align 4\n\t" | |
80 | ".word 1f,2\n\t" | |
81 | ".previous\n" | |
82 | "1:\n\t" | |
83 | "call __csum_partial_copy_sparc_generic\n\t" | |
84 | " st %8, [%%sp + 64]\n" | |
85 | : "=&r" (ret), "=&r" (d), "=&r" (l), "=&r" (s) | |
86 | : "0" (ret), "1" (d), "2" (l), "3" (s), "r" (err) | |
87 | : "o2", "o3", "o4", "o5", "o7", "g2", "g3", "g4", "g5", | |
88 | "cc", "memory"); | |
89 | return (__force __wsum)ret; | |
90 | } | |
91 | ||
2a89b674 AV |
92 | #define HAVE_CSUM_COPY_USER |
93 | ||
f5e706ad | 94 | static inline __wsum |
2a89b674 | 95 | csum_and_copy_to_user(const void *src, void __user *dst, int len, |
f5e706ad SR |
96 | __wsum sum, int *err) |
97 | { | |
96d4f267 | 98 | if (!access_ok(dst, len)) { |
f5e706ad SR |
99 | *err = -EFAULT; |
100 | return sum; | |
101 | } else { | |
102 | register unsigned long ret asm("o0") = (unsigned long)src; | |
103 | register char __user *d asm("o1") = dst; | |
104 | register int l asm("g1") = len; | |
105 | register __wsum s asm("g7") = sum; | |
106 | ||
107 | __asm__ __volatile__ ( | |
108 | ".section __ex_table,#alloc\n\t" | |
109 | ".align 4\n\t" | |
110 | ".word 1f,1\n\t" | |
111 | ".previous\n" | |
112 | "1:\n\t" | |
113 | "call __csum_partial_copy_sparc_generic\n\t" | |
114 | " st %8, [%%sp + 64]\n" | |
115 | : "=&r" (ret), "=&r" (d), "=&r" (l), "=&r" (s) | |
116 | : "0" (ret), "1" (d), "2" (l), "3" (s), "r" (err) | |
117 | : "o2", "o3", "o4", "o5", "o7", | |
118 | "g2", "g3", "g4", "g5", | |
119 | "cc", "memory"); | |
120 | return (__force __wsum)ret; | |
121 | } | |
122 | } | |
123 | ||
f5e706ad SR |
124 | /* ihl is always 5 or greater, almost always is 5, and iph is word aligned |
125 | * the majority of the time. | |
126 | */ | |
127 | static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl) | |
128 | { | |
129 | __sum16 sum; | |
130 | ||
131 | /* Note: We must read %2 before we touch %0 for the first time, | |
132 | * because GCC can legitimately use the same register for | |
133 | * both operands. | |
134 | */ | |
135 | __asm__ __volatile__("sub\t%2, 4, %%g4\n\t" | |
136 | "ld\t[%1 + 0x00], %0\n\t" | |
137 | "ld\t[%1 + 0x04], %%g2\n\t" | |
138 | "ld\t[%1 + 0x08], %%g3\n\t" | |
139 | "addcc\t%%g2, %0, %0\n\t" | |
140 | "addxcc\t%%g3, %0, %0\n\t" | |
141 | "ld\t[%1 + 0x0c], %%g2\n\t" | |
142 | "ld\t[%1 + 0x10], %%g3\n\t" | |
143 | "addxcc\t%%g2, %0, %0\n\t" | |
144 | "addx\t%0, %%g0, %0\n" | |
145 | "1:\taddcc\t%%g3, %0, %0\n\t" | |
146 | "add\t%1, 4, %1\n\t" | |
147 | "addxcc\t%0, %%g0, %0\n\t" | |
148 | "subcc\t%%g4, 1, %%g4\n\t" | |
149 | "be,a\t2f\n\t" | |
150 | "sll\t%0, 16, %%g2\n\t" | |
151 | "b\t1b\n\t" | |
152 | "ld\t[%1 + 0x10], %%g3\n" | |
153 | "2:\taddcc\t%0, %%g2, %%g2\n\t" | |
154 | "srl\t%%g2, 16, %0\n\t" | |
155 | "addx\t%0, %%g0, %0\n\t" | |
156 | "xnor\t%%g0, %0, %0" | |
157 | : "=r" (sum), "=&r" (iph) | |
158 | : "r" (ihl), "1" (iph) | |
159 | : "g2", "g3", "g4", "cc", "memory"); | |
160 | return sum; | |
161 | } | |
162 | ||
163 | /* Fold a partial checksum without adding pseudo headers. */ | |
164 | static inline __sum16 csum_fold(__wsum sum) | |
165 | { | |
166 | unsigned int tmp; | |
167 | ||
168 | __asm__ __volatile__("addcc\t%0, %1, %1\n\t" | |
169 | "srl\t%1, 16, %1\n\t" | |
170 | "addx\t%1, %%g0, %1\n\t" | |
171 | "xnor\t%%g0, %1, %0" | |
172 | : "=&r" (sum), "=r" (tmp) | |
173 | : "0" (sum), "1" ((__force u32)sum<<16) | |
174 | : "cc"); | |
175 | return (__force __sum16)sum; | |
176 | } | |
177 | ||
178 | static inline __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr, | |
01cfbad7 AD |
179 | __u32 len, __u8 proto, |
180 | __wsum sum) | |
f5e706ad SR |
181 | { |
182 | __asm__ __volatile__("addcc\t%1, %0, %0\n\t" | |
183 | "addxcc\t%2, %0, %0\n\t" | |
184 | "addxcc\t%3, %0, %0\n\t" | |
185 | "addx\t%0, %%g0, %0\n\t" | |
186 | : "=r" (sum), "=r" (saddr) | |
187 | : "r" (daddr), "r" (proto + len), "0" (sum), | |
188 | "1" (saddr) | |
189 | : "cc"); | |
190 | return sum; | |
191 | } | |
192 | ||
193 | /* | |
194 | * computes the checksum of the TCP/UDP pseudo-header | |
195 | * returns a 16-bit checksum, already complemented | |
196 | */ | |
197 | static inline __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr, | |
01cfbad7 AD |
198 | __u32 len, __u8 proto, |
199 | __wsum sum) | |
f5e706ad SR |
200 | { |
201 | return csum_fold(csum_tcpudp_nofold(saddr,daddr,len,proto,sum)); | |
202 | } | |
203 | ||
204 | #define _HAVE_ARCH_IPV6_CSUM | |
205 | ||
206 | static inline __sum16 csum_ipv6_magic(const struct in6_addr *saddr, | |
207 | const struct in6_addr *daddr, | |
1e940829 | 208 | __u32 len, __u8 proto, __wsum sum) |
f5e706ad SR |
209 | { |
210 | __asm__ __volatile__ ( | |
211 | "addcc %3, %4, %%g4\n\t" | |
212 | "addxcc %5, %%g4, %%g4\n\t" | |
213 | "ld [%2 + 0x0c], %%g2\n\t" | |
214 | "ld [%2 + 0x08], %%g3\n\t" | |
215 | "addxcc %%g2, %%g4, %%g4\n\t" | |
216 | "ld [%2 + 0x04], %%g2\n\t" | |
217 | "addxcc %%g3, %%g4, %%g4\n\t" | |
218 | "ld [%2 + 0x00], %%g3\n\t" | |
219 | "addxcc %%g2, %%g4, %%g4\n\t" | |
220 | "ld [%1 + 0x0c], %%g2\n\t" | |
221 | "addxcc %%g3, %%g4, %%g4\n\t" | |
222 | "ld [%1 + 0x08], %%g3\n\t" | |
223 | "addxcc %%g2, %%g4, %%g4\n\t" | |
224 | "ld [%1 + 0x04], %%g2\n\t" | |
225 | "addxcc %%g3, %%g4, %%g4\n\t" | |
226 | "ld [%1 + 0x00], %%g3\n\t" | |
227 | "addxcc %%g2, %%g4, %%g4\n\t" | |
228 | "addxcc %%g3, %%g4, %0\n\t" | |
229 | "addx 0, %0, %0\n" | |
230 | : "=&r" (sum) | |
231 | : "r" (saddr), "r" (daddr), | |
232 | "r"(htonl(len)), "r"(htonl(proto)), "r"(sum) | |
233 | : "g2", "g3", "g4", "cc"); | |
234 | ||
235 | return csum_fold(sum); | |
236 | } | |
237 | ||
238 | /* this routine is used for miscellaneous IP-like checksums, mainly in icmp.c */ | |
239 | static inline __sum16 ip_compute_csum(const void *buff, int len) | |
240 | { | |
241 | return csum_fold(csum_partial(buff, len, 0)); | |
242 | } | |
243 | ||
20fce54f TH |
244 | #define HAVE_ARCH_CSUM_ADD |
245 | static inline __wsum csum_add(__wsum csum, __wsum addend) | |
246 | { | |
247 | __asm__ __volatile__( | |
248 | "addcc %0, %1, %0\n" | |
249 | "addx %0, %%g0, %0" | |
250 | : "=r" (csum) | |
251 | : "r" (addend), "0" (csum)); | |
252 | ||
253 | return csum; | |
254 | } | |
255 | ||
f5e706ad | 256 | #endif /* !(__SPARC_CHECKSUM_H) */ |