Merge tag 'v3.3.7' into 3.3/master
[zen-stable.git] / arch / parisc / include / asm / checksum.h
blobc84b2fcb18a95ea1c643a6aea3595f2615721925
1 #ifndef _PARISC_CHECKSUM_H
2 #define _PARISC_CHECKSUM_H
4 #include <linux/in6.h>
6 /*
7 * computes the checksum of a memory block at buff, length len,
8 * and adds in "sum" (32-bit)
10 * returns a 32-bit number suitable for feeding into itself
11 * or csum_tcpudp_magic
13 * this function must be called with even lengths, except
14 * for the last fragment, which may be odd
16 * it's best to have buff aligned on a 32-bit boundary
18 extern __wsum csum_partial(const void *, int, __wsum);
21 * The same as csum_partial, but copies from src while it checksums.
23 * Here even more important to align src and dst on a 32-bit (or even
24 * better 64-bit) boundary
26 extern __wsum csum_partial_copy_nocheck(const void *, void *, int, __wsum);
29 * this is a new version of the above that records errors it finds in *errp,
30 * but continues and zeros the rest of the buffer.
32 extern __wsum csum_partial_copy_from_user(const void __user *src,
33 void *dst, int len, __wsum sum, int *errp);
36 * Optimized for IP headers, which always checksum on 4 octet boundaries.
38 * Written by Randolph Chung <tausq@debian.org>, and then mucked with by
39 * LaMont Jones <lamont@debian.org>
41 static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl)
43 unsigned int sum;
45 __asm__ __volatile__ (
46 " ldws,ma 4(%1), %0\n"
47 " addib,<= -4, %2, 2f\n"
48 "\n"
49 " ldws 4(%1), %%r20\n"
50 " ldws 8(%1), %%r21\n"
51 " add %0, %%r20, %0\n"
52 " ldws,ma 12(%1), %%r19\n"
53 " addc %0, %%r21, %0\n"
54 " addc %0, %%r19, %0\n"
55 "1: ldws,ma 4(%1), %%r19\n"
56 " addib,< 0, %2, 1b\n"
57 " addc %0, %%r19, %0\n"
58 "\n"
59 " extru %0, 31, 16, %%r20\n"
60 " extru %0, 15, 16, %%r21\n"
61 " addc %%r20, %%r21, %0\n"
62 " extru %0, 15, 16, %%r21\n"
63 " add %0, %%r21, %0\n"
64 " subi -1, %0, %0\n"
65 "2:\n"
66 : "=r" (sum), "=r" (iph), "=r" (ihl)
67 : "1" (iph), "2" (ihl)
68 : "r19", "r20", "r21", "memory");
70 return (__force __sum16)sum;
74 * Fold a partial checksum
76 static inline __sum16 csum_fold(__wsum csum)
78 u32 sum = (__force u32)csum;
79 /* add the swapped two 16-bit halves of sum,
80 a possible carry from adding the two 16-bit halves,
81 will carry from the lower half into the upper half,
82 giving us the correct sum in the upper half. */
83 sum += (sum << 16) + (sum >> 16);
84 return (__force __sum16)(~sum >> 16);
87 static inline __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr,
88 unsigned short len,
89 unsigned short proto,
90 __wsum sum)
92 __asm__(
93 " add %1, %0, %0\n"
94 " addc %2, %0, %0\n"
95 " addc %3, %0, %0\n"
96 " addc %%r0, %0, %0\n"
97 : "=r" (sum)
98 : "r" (daddr), "r"(saddr), "r"(proto+len), "0"(sum));
99 return sum;
103 * computes the checksum of the TCP/UDP pseudo-header
104 * returns a 16-bit checksum, already complemented
106 static inline __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr,
107 unsigned short len,
108 unsigned short proto,
109 __wsum sum)
111 return csum_fold(csum_tcpudp_nofold(saddr,daddr,len,proto,sum));
115 * this routine is used for miscellaneous IP-like checksums, mainly
116 * in icmp.c
118 static inline __sum16 ip_compute_csum(const void *buf, int len)
120 return csum_fold (csum_partial(buf, len, 0));
124 #define _HAVE_ARCH_IPV6_CSUM
125 static __inline__ __sum16 csum_ipv6_magic(const struct in6_addr *saddr,
126 const struct in6_addr *daddr,
127 __u32 len, unsigned short proto,
128 __wsum sum)
130 __asm__ __volatile__ (
132 #if BITS_PER_LONG > 32
135 ** We can execute two loads and two adds per cycle on PA 8000.
136 ** But add insn's get serialized waiting for the carry bit.
137 ** Try to keep 4 registers with "live" values ahead of the ALU.
140 " ldd,ma 8(%1), %%r19\n" /* get 1st saddr word */
141 " ldd,ma 8(%2), %%r20\n" /* get 1st daddr word */
142 " add %8, %3, %3\n"/* add 16-bit proto + len */
143 " add %%r19, %0, %0\n"
144 " ldd,ma 8(%1), %%r21\n" /* 2cd saddr */
145 " ldd,ma 8(%2), %%r22\n" /* 2cd daddr */
146 " add,dc %%r20, %0, %0\n"
147 " add,dc %%r21, %0, %0\n"
148 " add,dc %%r22, %0, %0\n"
149 " add,dc %3, %0, %0\n" /* fold in proto+len | carry bit */
150 " extrd,u %0, 31, 32, %%r19\n" /* copy upper half down */
151 " depdi 0, 31, 32, %0\n" /* clear upper half */
152 " add %%r19, %0, %0\n" /* fold into 32-bits */
153 " addc 0, %0, %0\n" /* add carry */
155 #else
158 ** For PA 1.x, the insn order doesn't matter as much.
159 ** Insn stream is serialized on the carry bit here too.
160 ** result from the previous operation (eg r0 + x)
163 " ldw,ma 4(%1), %%r19\n" /* get 1st saddr word */
164 " ldw,ma 4(%2), %%r20\n" /* get 1st daddr word */
165 " add %8, %3, %3\n" /* add 16-bit proto + len */
166 " add %%r19, %0, %0\n"
167 " ldw,ma 4(%1), %%r21\n" /* 2cd saddr */
168 " addc %%r20, %0, %0\n"
169 " ldw,ma 4(%2), %%r22\n" /* 2cd daddr */
170 " addc %%r21, %0, %0\n"
171 " ldw,ma 4(%1), %%r19\n" /* 3rd saddr */
172 " addc %%r22, %0, %0\n"
173 " ldw,ma 4(%2), %%r20\n" /* 3rd daddr */
174 " addc %%r19, %0, %0\n"
175 " ldw,ma 4(%1), %%r21\n" /* 4th saddr */
176 " addc %%r20, %0, %0\n"
177 " ldw,ma 4(%2), %%r22\n" /* 4th daddr */
178 " addc %%r21, %0, %0\n"
179 " addc %%r22, %0, %0\n"
180 " addc %3, %0, %0\n" /* fold in proto+len, catch carry */
182 #endif
183 : "=r" (sum), "=r" (saddr), "=r" (daddr), "=r" (len)
184 : "0" (sum), "1" (saddr), "2" (daddr), "3" (len), "r" (proto)
185 : "r19", "r20", "r21", "r22", "memory");
186 return csum_fold(sum);
190 * Copy and checksum to user
192 #define HAVE_CSUM_COPY_USER
193 static __inline__ __wsum csum_and_copy_to_user(const void *src,
194 void __user *dst,
195 int len, __wsum sum,
196 int *err_ptr)
198 /* code stolen from include/asm-mips64 */
199 sum = csum_partial(src, len, sum);
201 if (copy_to_user(dst, src, len)) {
202 *err_ptr = -EFAULT;
203 return (__force __wsum)-1;
206 return sum;
209 #endif