Merge tag 'v3.3.7' into 3.3/master
[zen-stable.git] / arch / sparc / include / asm / checksum_32.h
blobbdbda1453aa9f168339896d6880e3e2acd44182f
1 #ifndef __SPARC_CHECKSUM_H
2 #define __SPARC_CHECKSUM_H
4 /* checksum.h: IP/UDP/TCP checksum routines on the Sparc.
6 * Copyright(C) 1995 Linus Torvalds
7 * Copyright(C) 1995 Miguel de Icaza
8 * Copyright(C) 1996 David S. Miller
9 * Copyright(C) 1996 Eddie C. Dost
10 * Copyright(C) 1997 Jakub Jelinek
12 * derived from:
13 * Alpha checksum c-code
14 * ix86 inline assembly
15 * RFC1071 Computing the Internet Checksum
18 #include <linux/in6.h>
19 #include <asm/uaccess.h>
21 /* computes the checksum of a memory block at buff, length len,
22 * and adds in "sum" (32-bit)
24 * returns a 32-bit number suitable for feeding into itself
25 * or csum_tcpudp_magic
27 * this function must be called with even lengths, except
28 * for the last fragment, which may be odd
30 * it's best to have buff aligned on a 32-bit boundary
32 extern __wsum csum_partial(const void *buff, int len, __wsum sum);
34 /* the same as csum_partial, but copies from fs:src while it
35 * checksums
37 * here even more important to align src and dst on a 32-bit (or even
38 * better 64-bit) boundary
41 extern unsigned int __csum_partial_copy_sparc_generic (const unsigned char *, unsigned char *);
43 static inline __wsum
44 csum_partial_copy_nocheck(const void *src, void *dst, int len, __wsum sum)
46 register unsigned int ret asm("o0") = (unsigned int)src;
47 register char *d asm("o1") = dst;
48 register int l asm("g1") = len;
50 __asm__ __volatile__ (
51 "call __csum_partial_copy_sparc_generic\n\t"
52 " mov %6, %%g7\n"
53 : "=&r" (ret), "=&r" (d), "=&r" (l)
54 : "0" (ret), "1" (d), "2" (l), "r" (sum)
55 : "o2", "o3", "o4", "o5", "o7",
56 "g2", "g3", "g4", "g5", "g7",
57 "memory", "cc");
58 return (__force __wsum)ret;
61 static inline __wsum
62 csum_partial_copy_from_user(const void __user *src, void *dst, int len,
63 __wsum sum, int *err)
65 register unsigned long ret asm("o0") = (unsigned long)src;
66 register char *d asm("o1") = dst;
67 register int l asm("g1") = len;
68 register __wsum s asm("g7") = sum;
70 __asm__ __volatile__ (
71 ".section __ex_table,#alloc\n\t"
72 ".align 4\n\t"
73 ".word 1f,2\n\t"
74 ".previous\n"
75 "1:\n\t"
76 "call __csum_partial_copy_sparc_generic\n\t"
77 " st %8, [%%sp + 64]\n"
78 : "=&r" (ret), "=&r" (d), "=&r" (l), "=&r" (s)
79 : "0" (ret), "1" (d), "2" (l), "3" (s), "r" (err)
80 : "o2", "o3", "o4", "o5", "o7", "g2", "g3", "g4", "g5",
81 "cc", "memory");
82 return (__force __wsum)ret;
85 static inline __wsum
86 csum_partial_copy_to_user(const void *src, void __user *dst, int len,
87 __wsum sum, int *err)
89 if (!access_ok (VERIFY_WRITE, dst, len)) {
90 *err = -EFAULT;
91 return sum;
92 } else {
93 register unsigned long ret asm("o0") = (unsigned long)src;
94 register char __user *d asm("o1") = dst;
95 register int l asm("g1") = len;
96 register __wsum s asm("g7") = sum;
98 __asm__ __volatile__ (
99 ".section __ex_table,#alloc\n\t"
100 ".align 4\n\t"
101 ".word 1f,1\n\t"
102 ".previous\n"
103 "1:\n\t"
104 "call __csum_partial_copy_sparc_generic\n\t"
105 " st %8, [%%sp + 64]\n"
106 : "=&r" (ret), "=&r" (d), "=&r" (l), "=&r" (s)
107 : "0" (ret), "1" (d), "2" (l), "3" (s), "r" (err)
108 : "o2", "o3", "o4", "o5", "o7",
109 "g2", "g3", "g4", "g5",
110 "cc", "memory");
111 return (__force __wsum)ret;
115 #define HAVE_CSUM_COPY_USER
116 #define csum_and_copy_to_user csum_partial_copy_to_user
118 /* ihl is always 5 or greater, almost always is 5, and iph is word aligned
119 * the majority of the time.
121 static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl)
123 __sum16 sum;
125 /* Note: We must read %2 before we touch %0 for the first time,
126 * because GCC can legitimately use the same register for
127 * both operands.
129 __asm__ __volatile__("sub\t%2, 4, %%g4\n\t"
130 "ld\t[%1 + 0x00], %0\n\t"
131 "ld\t[%1 + 0x04], %%g2\n\t"
132 "ld\t[%1 + 0x08], %%g3\n\t"
133 "addcc\t%%g2, %0, %0\n\t"
134 "addxcc\t%%g3, %0, %0\n\t"
135 "ld\t[%1 + 0x0c], %%g2\n\t"
136 "ld\t[%1 + 0x10], %%g3\n\t"
137 "addxcc\t%%g2, %0, %0\n\t"
138 "addx\t%0, %%g0, %0\n"
139 "1:\taddcc\t%%g3, %0, %0\n\t"
140 "add\t%1, 4, %1\n\t"
141 "addxcc\t%0, %%g0, %0\n\t"
142 "subcc\t%%g4, 1, %%g4\n\t"
143 "be,a\t2f\n\t"
144 "sll\t%0, 16, %%g2\n\t"
145 "b\t1b\n\t"
146 "ld\t[%1 + 0x10], %%g3\n"
147 "2:\taddcc\t%0, %%g2, %%g2\n\t"
148 "srl\t%%g2, 16, %0\n\t"
149 "addx\t%0, %%g0, %0\n\t"
150 "xnor\t%%g0, %0, %0"
151 : "=r" (sum), "=&r" (iph)
152 : "r" (ihl), "1" (iph)
153 : "g2", "g3", "g4", "cc", "memory");
154 return sum;
157 /* Fold a partial checksum without adding pseudo headers. */
158 static inline __sum16 csum_fold(__wsum sum)
160 unsigned int tmp;
162 __asm__ __volatile__("addcc\t%0, %1, %1\n\t"
163 "srl\t%1, 16, %1\n\t"
164 "addx\t%1, %%g0, %1\n\t"
165 "xnor\t%%g0, %1, %0"
166 : "=&r" (sum), "=r" (tmp)
167 : "0" (sum), "1" ((__force u32)sum<<16)
168 : "cc");
169 return (__force __sum16)sum;
172 static inline __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr,
173 unsigned short len,
174 unsigned short proto,
175 __wsum sum)
177 __asm__ __volatile__("addcc\t%1, %0, %0\n\t"
178 "addxcc\t%2, %0, %0\n\t"
179 "addxcc\t%3, %0, %0\n\t"
180 "addx\t%0, %%g0, %0\n\t"
181 : "=r" (sum), "=r" (saddr)
182 : "r" (daddr), "r" (proto + len), "0" (sum),
183 "1" (saddr)
184 : "cc");
185 return sum;
189 * computes the checksum of the TCP/UDP pseudo-header
190 * returns a 16-bit checksum, already complemented
192 static inline __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr,
193 unsigned short len,
194 unsigned short proto,
195 __wsum sum)
197 return csum_fold(csum_tcpudp_nofold(saddr,daddr,len,proto,sum));
200 #define _HAVE_ARCH_IPV6_CSUM
202 static inline __sum16 csum_ipv6_magic(const struct in6_addr *saddr,
203 const struct in6_addr *daddr,
204 __u32 len, unsigned short proto,
205 __wsum sum)
207 __asm__ __volatile__ (
208 "addcc %3, %4, %%g4\n\t"
209 "addxcc %5, %%g4, %%g4\n\t"
210 "ld [%2 + 0x0c], %%g2\n\t"
211 "ld [%2 + 0x08], %%g3\n\t"
212 "addxcc %%g2, %%g4, %%g4\n\t"
213 "ld [%2 + 0x04], %%g2\n\t"
214 "addxcc %%g3, %%g4, %%g4\n\t"
215 "ld [%2 + 0x00], %%g3\n\t"
216 "addxcc %%g2, %%g4, %%g4\n\t"
217 "ld [%1 + 0x0c], %%g2\n\t"
218 "addxcc %%g3, %%g4, %%g4\n\t"
219 "ld [%1 + 0x08], %%g3\n\t"
220 "addxcc %%g2, %%g4, %%g4\n\t"
221 "ld [%1 + 0x04], %%g2\n\t"
222 "addxcc %%g3, %%g4, %%g4\n\t"
223 "ld [%1 + 0x00], %%g3\n\t"
224 "addxcc %%g2, %%g4, %%g4\n\t"
225 "addxcc %%g3, %%g4, %0\n\t"
226 "addx 0, %0, %0\n"
227 : "=&r" (sum)
228 : "r" (saddr), "r" (daddr),
229 "r"(htonl(len)), "r"(htonl(proto)), "r"(sum)
230 : "g2", "g3", "g4", "cc");
232 return csum_fold(sum);
235 /* this routine is used for miscellaneous IP-like checksums, mainly in icmp.c */
236 static inline __sum16 ip_compute_csum(const void *buff, int len)
238 return csum_fold(csum_partial(buff, len, 0));
241 #endif /* !(__SPARC_CHECKSUM_H) */