xref: /original-bsd/sys/vax/vax/in_cksum.c (revision 0fc6f013)
1 /*
2  * Copyright (c) 1982 Regents of the University of California.
3  * All rights reserved.  The Berkeley software License Agreement
4  * specifies the terms and conditions for redistribution.
5  *
6  *	@(#)in_cksum.c	6.4 (Berkeley) 06/08/85
7  */
8 
9 #include "types.h"
10 #include "mbuf.h"
11 #include "../netinet/in.h"
12 #include "../netinet/in_systm.h"
13 
14 /*
15  * Checksum routine for Internet Protocol family headers (VAX Version).
16  *
17  * This routine is very heavily used in the network
18  * code and should be modified for each CPU to be as fast as possible.
19  */
20 
21 in_cksum(m, len)
22 	register struct mbuf *m;
23 	register int len;
24 {
25 	register u_short *w;		/* on vax, known to be r9 */
26 	register int sum = 0;		/* on vax, known to be r8 */
27 	register int mlen = 0;
28 
29 	for (;;) {
30 		/*
31 		 * Each trip around loop adds in
32 		 * word from one mbuf segment.
33 		 */
34 		w = mtod(m, u_short *);
35 		if (mlen == -1) {
36 			/*
37 			 * There is a byte left from the last segment;
38 			 * add it into the checksum.  Don't have to worry
39 			 * about a carry-out here because we make sure
40 			 * that high part of (32 bit) sum is small below.
41 			 */
42 			sum += *(u_char *)w << 8;
43 			w = (u_short *)((char *)w + 1);
44 			mlen = m->m_len - 1;
45 			len--;
46 		} else
47 			mlen = m->m_len;
48 		m = m->m_next;
49 		if (len < mlen)
50 			mlen = len;
51 		len -= mlen;
52 		/*
53 		 * Force to long boundary so we do longword aligned
54 		 * memory operations.  It is too hard to do byte
55 		 * adjustment, do only word adjustment.
56 		 */
57 		if (((int)w&0x2) && mlen >= 2) {
58 			sum += *w++;
59 			mlen -= 2;
60 		}
61 		/*
62 		 * Do as much of the checksum as possible 32 bits at at time.
63 		 * In fact, this loop is unrolled to make overhead from
64 		 * branches &c small.
65 		 *
66 		 * We can do a 16 bit ones complement sum 32 bits at a time
67 		 * because the 32 bit register is acting as two 16 bit
68 		 * registers for adding, with carries from the low added
69 		 * into the high (by normal carry-chaining) and carries
70 		 * from the high carried into the low on the next word
71 		 * by use of the adwc instruction.  This lets us run
72 		 * this loop at almost memory speed.
73 		 *
74 		 * Here there is the danger of high order carry out, and
75 		 * we carefully use adwc.
76 		 */
77 		while ((mlen -= 32) >= 0) {
78 #undef ADD
79 #ifdef unneeded		 /* The loop construct clears carry for us... */
80 			asm("bicpsr $1");		/* clears carry */
81 #endif
82 #define ADD		asm("adwc (r9)+,r8;");
83 			ADD; ADD; ADD; ADD; ADD; ADD; ADD; ADD;
84 			asm("adwc $0,r8");
85 		}
86 		mlen += 32;
87 		while ((mlen -= 8) >= 0) {
88 #ifdef unneeded		 /* The loop construct clears carry for us... */
89 			asm("bicpsr $1");		/* clears carry */
90 #endif
91 			ADD; ADD;
92 			asm("adwc $0,r8");
93 		}
94 		mlen += 8;
95 		/*
96 		 * Now eliminate the possibility of carry-out's by
97 		 * folding back to a 16 bit number (adding high and
98 		 * low parts together.)  Then mop up trailing words
99 		 * and maybe an odd byte.
100 		 */
101 		{ asm("ashl $-16,r8,r0; addw2 r0,r8");
102 		  asm("adwc $0,r8; movzwl r8,r8"); }
103 		while ((mlen -= 2) >= 0) {
104 			asm("movzwl (r9)+,r0; addl2 r0,r8");
105 		}
106 		if (mlen == -1) {
107 			sum += *(u_char *)w;
108 		}
109 		if (len == 0)
110 			break;
111 		/*
112 		 * Locate the next block with some data.
113 		 * If there is a word split across a boundary we
114 		 * will wrap to the top with mlen == -1 and
115 		 * then add it in shifted appropriately.
116 		 */
117 		for (;;) {
118 			if (m == 0) {
119 				printf("cksum: out of data\n");
120 				goto done;
121 			}
122 			if (m->m_len)
123 				break;
124 			m = m->m_next;
125 		}
126 	}
127 done:
128 	/*
129 	 * Add together high and low parts of sum
130 	 * and carry to get cksum.
131 	 * Have to be careful to not drop the last
132 	 * carry here.
133 	 */
134 	{ asm("ashl $-16,r8,r0; addw2 r0,r8; adwc $0,r8");
135 	  asm("mcoml r8,r8; movzwl r8,r8"); }
136 	return (sum);
137 }
138