Linux 2.6.12-rc6
[cris-mirror.git] / include / asm-parisc / checksum.h
blob229cb56fdb7a01713188d52f0a47207a6e54fbe5
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 unsigned int csum_partial(const unsigned char *, int, unsigned int);
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 unsigned int csum_partial_copy_nocheck(const unsigned char *, unsigned char *,
27 int, unsigned int);
30 * this is a new version of the above that records errors it finds in *errp,
31 * but continues and zeros the rest of the buffer.
33 extern unsigned int csum_partial_copy_from_user(const unsigned char __user *src,
34 unsigned char *dst, int len, unsigned int sum, int *errp);
37 * Optimized for IP headers, which always checksum on 4 octet boundaries.
39 * Written by Randolph Chung <tausq@debian.org>, and then mucked with by
40 * LaMont Jones <lamont@debian.org>
42 static inline unsigned short ip_fast_csum(unsigned char * iph,
43 unsigned int ihl) {
44 unsigned int sum;
47 __asm__ __volatile__ (
48 " ldws,ma 4(%1), %0\n"
49 " addib,<= -4, %2, 2f\n"
50 "\n"
51 " ldws 4(%1), %%r20\n"
52 " ldws 8(%1), %%r21\n"
53 " add %0, %%r20, %0\n"
54 " ldws,ma 12(%1), %%r19\n"
55 " addc %0, %%r21, %0\n"
56 " addc %0, %%r19, %0\n"
57 "1: ldws,ma 4(%1), %%r19\n"
58 " addib,< 0, %2, 1b\n"
59 " addc %0, %%r19, %0\n"
60 "\n"
61 " extru %0, 31, 16, %%r20\n"
62 " extru %0, 15, 16, %%r21\n"
63 " addc %%r20, %%r21, %0\n"
64 " extru %0, 15, 16, %%r21\n"
65 " add %0, %%r21, %0\n"
66 " subi -1, %0, %0\n"
67 "2:\n"
68 : "=r" (sum), "=r" (iph), "=r" (ihl)
69 : "1" (iph), "2" (ihl)
70 : "r19", "r20", "r21" );
72 return(sum);
76 * Fold a partial checksum
78 static inline unsigned int csum_fold(unsigned int sum)
80 /* add the swapped two 16-bit halves of sum,
81 a possible carry from adding the two 16-bit halves,
82 will carry from the lower half into the upper half,
83 giving us the correct sum in the upper half. */
84 sum += (sum << 16) + (sum >> 16);
85 return (~sum) >> 16;
88 static inline unsigned long csum_tcpudp_nofold(unsigned long saddr,
89 unsigned long daddr,
90 unsigned short len,
91 unsigned short proto,
92 unsigned int sum)
94 __asm__(
95 " add %1, %0, %0\n"
96 " addc %2, %0, %0\n"
97 " addc %3, %0, %0\n"
98 " addc %%r0, %0, %0\n"
99 : "=r" (sum)
100 : "r" (daddr), "r"(saddr), "r"((proto<<16)+len), "0"(sum));
101 return sum;
105 * computes the checksum of the TCP/UDP pseudo-header
106 * returns a 16-bit checksum, already complemented
108 static inline unsigned short int csum_tcpudp_magic(unsigned long saddr,
109 unsigned long daddr,
110 unsigned short len,
111 unsigned short proto,
112 unsigned int sum)
114 return csum_fold(csum_tcpudp_nofold(saddr,daddr,len,proto,sum));
118 * this routine is used for miscellaneous IP-like checksums, mainly
119 * in icmp.c
121 static inline unsigned short ip_compute_csum(unsigned char * buf, int len) {
122 return csum_fold (csum_partial(buf, len, 0));
126 #define _HAVE_ARCH_IPV6_CSUM
127 static __inline__ unsigned short int csum_ipv6_magic(struct in6_addr *saddr,
128 struct in6_addr *daddr,
129 __u16 len,
130 unsigned short proto,
131 unsigned int sum)
133 __asm__ __volatile__ (
135 #if BITS_PER_LONG > 32
138 ** We can execute two loads and two adds per cycle on PA 8000.
139 ** But add insn's get serialized waiting for the carry bit.
140 ** Try to keep 4 registers with "live" values ahead of the ALU.
143 " ldd,ma 8(%1), %%r19\n" /* get 1st saddr word */
144 " ldd,ma 8(%2), %%r20\n" /* get 1st daddr word */
145 " add %8, %3, %3\n"/* add 16-bit proto + len */
146 " add %%r19, %0, %0\n"
147 " ldd,ma 8(%1), %%r21\n" /* 2cd saddr */
148 " ldd,ma 8(%2), %%r22\n" /* 2cd daddr */
149 " add,dc %%r20, %0, %0\n"
150 " add,dc %%r21, %0, %0\n"
151 " add,dc %%r22, %0, %0\n"
152 " add,dc %3, %0, %0\n" /* fold in proto+len | carry bit */
153 " extrd,u %0, 31, 32, %%r19\n" /* copy upper half down */
154 " depdi 0, 31, 32, %0\n" /* clear upper half */
155 " add %%r19, %0, %0\n" /* fold into 32-bits */
156 " addc 0, %0, %0\n" /* add carry */
158 #else
161 ** For PA 1.x, the insn order doesn't matter as much.
162 ** Insn stream is serialized on the carry bit here too.
163 ** result from the previous operation (eg r0 + x)
166 " ldw,ma 4(%1), %%r19\n" /* get 1st saddr word */
167 " ldw,ma 4(%2), %%r20\n" /* get 1st daddr word */
168 " add %8, %3, %3\n" /* add 16-bit proto + len */
169 " add %%r19, %0, %0\n"
170 " ldw,ma 4(%1), %%r21\n" /* 2cd saddr */
171 " addc %%r20, %0, %0\n"
172 " ldw,ma 4(%2), %%r22\n" /* 2cd daddr */
173 " addc %%r21, %0, %0\n"
174 " ldw,ma 4(%1), %%r19\n" /* 3rd saddr */
175 " addc %%r22, %0, %0\n"
176 " ldw,ma 4(%2), %%r20\n" /* 3rd daddr */
177 " addc %%r19, %0, %0\n"
178 " ldw,ma 4(%1), %%r21\n" /* 4th saddr */
179 " addc %%r20, %0, %0\n"
180 " ldw,ma 4(%2), %%r22\n" /* 4th daddr */
181 " addc %%r21, %0, %0\n"
182 " addc %%r22, %0, %0\n"
183 " addc %3, %0, %0\n" /* fold in proto+len, catch carry */
185 #endif
186 : "=r" (sum), "=r" (saddr), "=r" (daddr), "=r" (len)
187 : "0" (sum), "1" (saddr), "2" (daddr), "3" (len), "r" (proto)
188 : "r19", "r20", "r21", "r22");
189 return csum_fold(sum);
193 * Copy and checksum to user
195 #define HAVE_CSUM_COPY_USER
196 static __inline__ unsigned int csum_and_copy_to_user (const unsigned char *src,
197 unsigned char __user *dst,
198 int len, int sum,
199 int *err_ptr)
201 /* code stolen from include/asm-mips64 */
202 sum = csum_partial(src, len, sum);
204 if (copy_to_user(dst, src, len)) {
205 *err_ptr = -EFAULT;
206 return -1;
209 return sum;
212 #endif