blob: 471f2af2b16ae58d91a214402da3d416d1f0e435 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#ifndef _S390_CHECKSUM_H
2#define _S390_CHECKSUM_H
3
4/*
5 * include/asm-s390/checksum.h
6 * S390 fast network checksum routines
7 * see also arch/S390/lib/checksum.c
8 *
9 * S390 version
10 * Copyright (C) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
11 * Author(s): Ulrich Hild (first version)
12 * Martin Schwidefsky (heavily optimized CKSM version)
13 * D.J. Barrow (third attempt)
14 */
15
16#include <asm/uaccess.h>
17
18/*
19 * computes the checksum of a memory block at buff, length len,
20 * and adds in "sum" (32-bit)
21 *
22 * returns a 32-bit number suitable for feeding into itself
23 * or csum_tcpudp_magic
24 *
25 * this function must be called with even lengths, except
26 * for the last fragment, which may be odd
27 *
28 * it's best to have buff aligned on a 32-bit boundary
29 */
30static inline unsigned int
31csum_partial(const unsigned char * buff, int len, unsigned int sum)
32{
33 /*
34 * Experiments with ethernet and slip connections show that buf
35 * is aligned on either a 2-byte or 4-byte boundary.
36 */
37#ifndef __s390x__
38 register_pair rp;
39
40 rp.subreg.even = (unsigned long) buff;
41 rp.subreg.odd = (unsigned long) len;
42 __asm__ __volatile__ (
43 "0: cksm %0,%1\n" /* do checksum on longs */
44 " jo 0b\n"
45 : "+&d" (sum), "+&a" (rp) : : "cc", "memory" );
46#else /* __s390x__ */
47 __asm__ __volatile__ (
48 " lgr 2,%1\n" /* address in gpr 2 */
49 " lgfr 3,%2\n" /* length in gpr 3 */
50 "0: cksm %0,2\n" /* do checksum on longs */
51 " jo 0b\n"
52 : "+&d" (sum)
53 : "d" (buff), "d" (len)
54 : "cc", "memory", "2", "3" );
55#endif /* __s390x__ */
56 return sum;
57}
58
59/*
60 * csum_partial as an inline function
61 */
62static inline unsigned int
63csum_partial_inline(const unsigned char * buff, int len, unsigned int sum)
64{
65#ifndef __s390x__
66 register_pair rp;
67
68 rp.subreg.even = (unsigned long) buff;
69 rp.subreg.odd = (unsigned long) len;
70 __asm__ __volatile__ (
71 "0: cksm %0,%1\n" /* do checksum on longs */
72 " jo 0b\n"
73 : "+&d" (sum), "+&a" (rp) : : "cc", "memory" );
74#else /* __s390x__ */
75 __asm__ __volatile__ (
76 " lgr 2,%1\n" /* address in gpr 2 */
77 " lgfr 3,%2\n" /* length in gpr 3 */
78 "0: cksm %0,2\n" /* do checksum on longs */
79 " jo 0b\n"
80 : "+&d" (sum)
81 : "d" (buff), "d" (len)
82 : "cc", "memory", "2", "3" );
83#endif /* __s390x__ */
84 return sum;
85}
86
87/*
88 * the same as csum_partial_copy, but copies from user space.
89 *
90 * here even more important to align src and dst on a 32-bit (or even
91 * better 64-bit) boundary
92 *
93 * Copy from userspace and compute checksum. If we catch an exception
94 * then zero the rest of the buffer.
95 */
96static inline unsigned int
97csum_partial_copy_from_user(const char __user *src, char *dst,
98 int len, unsigned int sum,
99 int *err_ptr)
100{
101 int missing;
102
103 missing = copy_from_user(dst, src, len);
104 if (missing) {
105 memset(dst + len - missing, 0, missing);
106 *err_ptr = -EFAULT;
107 }
108
109 return csum_partial(dst, len, sum);
110}
111
112
113static inline unsigned int
114csum_partial_copy_nocheck (const char *src, char *dst, int len, unsigned int sum)
115{
116 memcpy(dst,src,len);
117 return csum_partial_inline(dst, len, sum);
118}
119
120/*
121 * Fold a partial checksum without adding pseudo headers
122 */
123static inline unsigned short
124csum_fold(unsigned int sum)
125{
126#ifndef __s390x__
127 register_pair rp;
128
129 __asm__ __volatile__ (
130 " slr %N1,%N1\n" /* %0 = H L */
131 " lr %1,%0\n" /* %0 = H L, %1 = H L 0 0 */
132 " srdl %1,16\n" /* %0 = H L, %1 = 0 H L 0 */
133 " alr %1,%N1\n" /* %0 = H L, %1 = L H L 0 */
134 " alr %0,%1\n" /* %0 = H+L+C L+H */
135 " srl %0,16\n" /* %0 = H+L+C */
136 : "+&d" (sum), "=d" (rp) : : "cc" );
137#else /* __s390x__ */
138 __asm__ __volatile__ (
139 " sr 3,3\n" /* %0 = H*65536 + L */
140 " lr 2,%0\n" /* %0 = H L, R2/R3 = H L / 0 0 */
141 " srdl 2,16\n" /* %0 = H L, R2/R3 = 0 H / L 0 */
142 " alr 2,3\n" /* %0 = H L, R2/R3 = L H / L 0 */
143 " alr %0,2\n" /* %0 = H+L+C L+H */
144 " srl %0,16\n" /* %0 = H+L+C */
145 : "+&d" (sum) : : "cc", "2", "3");
146#endif /* __s390x__ */
147 return ((unsigned short) ~sum);
148}
149
150/*
151 * This is a version of ip_compute_csum() optimized for IP headers,
152 * which always checksum on 4 octet boundaries.
153 *
154 */
155static inline unsigned short
156ip_fast_csum(unsigned char *iph, unsigned int ihl)
157{
158 unsigned long sum;
159#ifndef __s390x__
160 register_pair rp;
161
162 rp.subreg.even = (unsigned long) iph;
163 rp.subreg.odd = (unsigned long) ihl*4;
164 __asm__ __volatile__ (
165 " sr %0,%0\n" /* set sum to zero */
166 "0: cksm %0,%1\n" /* do checksum on longs */
167 " jo 0b\n"
168 : "=&d" (sum), "+&a" (rp) : : "cc", "memory" );
169#else /* __s390x__ */
170 __asm__ __volatile__ (
171 " slgr %0,%0\n" /* set sum to zero */
172 " lgr 2,%1\n" /* address in gpr 2 */
173 " lgfr 3,%2\n" /* length in gpr 3 */
174 "0: cksm %0,2\n" /* do checksum on ints */
175 " jo 0b\n"
176 : "=&d" (sum)
177 : "d" (iph), "d" (ihl*4)
178 : "cc", "memory", "2", "3" );
179#endif /* __s390x__ */
180 return csum_fold(sum);
181}
182
183/*
184 * computes the checksum of the TCP/UDP pseudo-header
185 * returns a 32-bit checksum
186 */
187static inline unsigned int
188csum_tcpudp_nofold(unsigned long saddr, unsigned long daddr,
189 unsigned short len, unsigned short proto,
190 unsigned int sum)
191{
192#ifndef __s390x__
193 __asm__ __volatile__ (
194 " alr %0,%1\n" /* sum += saddr */
195 " brc 12,0f\n"
196 " ahi %0,1\n" /* add carry */
197 "0:"
198 : "+&d" (sum) : "d" (saddr) : "cc" );
199 __asm__ __volatile__ (
200 " alr %0,%1\n" /* sum += daddr */
201 " brc 12,1f\n"
202 " ahi %0,1\n" /* add carry */
203 "1:"
204 : "+&d" (sum) : "d" (daddr) : "cc" );
205 __asm__ __volatile__ (
206 " alr %0,%1\n" /* sum += (len<<16) + (proto<<8) */
207 " brc 12,2f\n"
208 " ahi %0,1\n" /* add carry */
209 "2:"
210 : "+&d" (sum)
211 : "d" (((unsigned int) len<<16) + (unsigned int) proto)
212 : "cc" );
213#else /* __s390x__ */
214 __asm__ __volatile__ (
215 " lgfr %0,%0\n"
216 " algr %0,%1\n" /* sum += saddr */
217 " brc 12,0f\n"
218 " aghi %0,1\n" /* add carry */
219 "0: algr %0,%2\n" /* sum += daddr */
220 " brc 12,1f\n"
221 " aghi %0,1\n" /* add carry */
222 "1: algfr %0,%3\n" /* sum += (len<<16) + proto */
223 " brc 12,2f\n"
224 " aghi %0,1\n" /* add carry */
225 "2: srlg 0,%0,32\n"
226 " alr %0,0\n" /* fold to 32 bits */
227 " brc 12,3f\n"
228 " ahi %0,1\n" /* add carry */
229 "3: llgfr %0,%0"
230 : "+&d" (sum)
231 : "d" (saddr), "d" (daddr),
232 "d" (((unsigned int) len<<16) + (unsigned int) proto)
233 : "cc", "0" );
234#endif /* __s390x__ */
235 return sum;
236}
237
238/*
239 * computes the checksum of the TCP/UDP pseudo-header
240 * returns a 16-bit checksum, already complemented
241 */
242
243static inline unsigned short int
244csum_tcpudp_magic(unsigned long saddr, unsigned long daddr,
245 unsigned short len, unsigned short proto,
246 unsigned int sum)
247{
248 return csum_fold(csum_tcpudp_nofold(saddr,daddr,len,proto,sum));
249}
250
251/*
252 * this routine is used for miscellaneous IP-like checksums, mainly
253 * in icmp.c
254 */
255
256static inline unsigned short
257ip_compute_csum(unsigned char * buff, int len)
258{
259 return csum_fold(csum_partial(buff, len, 0));
260}
261
262#endif /* _S390_CHECKSUM_H */
263
264