Added i386 specific version of in_cksum.c and restructured the main

file to switch out to CPU specific implementations.

3 files changed, 515 insertions, 151 deletions

diff --git a/c/src/exec/libnetworking/netinet/in_cksum.c b/c/src/exec/libnetworking/netinet/in_cksum.c
index a2889f878a..9cffad3221 100644
--- a/c/src/exec/libnetworking/netinet/in_cksum.c
+++ b/c/src/exec/libnetworking/netinet/in_cksum.c
@@ -37,165 +37,20 @@
 #include <sys/param.h>
 #include <sys/mbuf.h>
 
-#if (defined(__GNUC__) && (defined(__mc68000__) || defined(__m68k__)))
-
-#define REDUCE { sum = (sum & 0xFFFF) + (sum >> 16); if (sum > 0xFFFF) sum -= 0xFFFF; }
-
 /*
- * Motorola 68k version of Internet Protocol Checksum routine
- * W. Eric Norum
- * Saskatchewan Accelerator Laboratory
- * August, 1998
+ *  Try to use a CPU specific version, then punt to the portable C one.
  */
-int
-in_cksum(m, len)
-	struct mbuf *m;
-	int len;
-{
-	unsigned short *w;
-	unsigned long sum = 0;
-	int mlen = 0;
-	int byte_swapped = 0;
-	union {
-		char	c[2];
-		u_short	s;
-	} s_util;
 
-	for ( ; m && len ; m = m->m_next) {
-		if (m->m_len == 0)
-			continue;
-		w = mtod(m, u_short *);
-		if (mlen == -1) {
-			/*
-			 * The first byte of this mbuf is the continuation
-			 * of a word spanning between this mbuf and the
-			 * last mbuf.
-			 *
-			 * s_util.c[0] is already saved when scanning previous
-			 * mbuf.
-			 */
-			s_util.c[1] = *(char *)w;
-			sum += s_util.s;
-			w = (u_short *)((char *)w + 1);
-			mlen = m->m_len - 1;
-			len--;
-		} else
-			mlen = m->m_len;
-		if (len < mlen)
-			mlen = len;
-		len -= mlen;
-
-		/*
-		 * Force to longword boundary.
-		 */
-		if (3 & (int)w) {
-			REDUCE;
-			if ((1 & (int) w) && (mlen > 0)) {
-				sum <<= 8;
-				s_util.c[0] = *(u_char *)w;
-				w = (u_short *)((char *)w + 1);
-				mlen--;
-				byte_swapped = 1;
-			}
-			if ((2 & (int) w) && (mlen >= 2)) {
-				sum += *w++;
-				mlen -= 2;
-			}
-		}
+#if (defined(__GNUC__) && (defined(__mc68000__) || defined(__m68k__)))
 
-		/*
-		 * Sum all the longwords in the buffer.
-		 * See RFC 1071 -- Computing the Internet Checksum.
-		 * It should work for all 68k family members.
-		 */
-		{
-		unsigned long tcnt = mlen, t1;
-		__asm__ volatile (
-		"movel   %2,%3\n\t"
-		"lsrl    #6,%2       | count/64 = # loop traversals\n\t"
-		"andl    #0x3c,%3    | Then find fractions of a chunk\n\t"
-		"negl    %3\n\t      | Each long uses 4 instruction bytes\n\t"
-		"andi    #0xf,%%cc   | Clear X (extended carry flag)\n\t"
-		"jmp     %%pc@(lcsum2_lbl-.-2:b,%3)  | Jump into loop\n"
-		"lcsum1_lbl:     | Begin inner loop...\n\t"
-		"movel   %1@+,%3     |  0: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  1: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  2: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  3: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  4: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  5: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  6: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  7: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  8: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  9: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  A: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  B: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  C: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  D: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  E: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n\t"
-		"movel   %1@+,%3     |  F: Fetch 32-bit word\n\t"
-		"addxl   %3,%0       |    Add word + previous carry\n"
-		"lcsum2_lbl:\n\tdbf    %2,lcsum1_lbl   | (NB- dbra doesn't affect X)\n\t"
-		"movel    %0,%3         | Fold 32 bit sum to 16 bits\n\t"
-		"swap    %3             | (NB- swap doesn't affect X)\n\t"
-		"addxw   %3,%0          |\n\t"
-		"moveq   #0,%3          | Add in last carry\n\t"
-		"addxw   %3,%0          |\n\t"
-		"andl    #0xffff,%0     | Mask to 16-bit sum\n" :
-			"=d" (sum), "=a" (w), "=d" (tcnt) , "=d" (t1) :
-			"0" (sum), "1" (w), "2" (tcnt) :
-			"cc", "memory");
-		}
-		mlen &= 3;
+#include "in_cksum_m68k.c"
 
-		/*
-		 * Soak up the last 1, 2 or 3 bytes
-		 */
-		while ((mlen -= 2) >= 0)
-			sum += *w++;
-		if (byte_swapped) {
-			REDUCE;
-			sum <<= 8;
-			byte_swapped = 0;
-			if (mlen == -1) {
-				s_util.c[1] = *(char *)w;
-				sum += s_util.s;
-				mlen = 0;
-			} else
-				mlen = -1;
-		} else if (mlen == -1)
-			s_util.c[0] = *(char *)w;
-	}
-	if (len)
-		sum = 0xDEAD;
-	if (mlen == -1) {
-		/* The last mbuf has odd # of bytes. Follow the
-		   standard (the odd byte may be shifted left by 8 bits
-		   or not as determined by endian-ness of the machine) */
-		s_util.c[1] = 0;
-		sum += s_util.s;
-	}
-	REDUCE;
-	return (~sum & 0xffff);
-}
+#elif (defined(__GNUC__) && defined(__i386__))
 
+#include "in_cksum_i386.c"
 
 #else
+
 /*
  * Checksum routine for Internet Protocol family headers (Portable Version).
  *
diff --git a/c/src/exec/libnetworking/netinet/in_cksum_i386.c b/c/src/exec/libnetworking/netinet/in_cksum_i386.c
new file mode 100644
index 0000000000..2057ab3fd6
--- /dev/null
+++ b/c/src/exec/libnetworking/netinet/in_cksum_i386.c
@@ -0,0 +1,198 @@
+/*
+ * Checksum routine for Internet Protocol family headers.
+ *
+ * This routine is very heavily used in the network
+ * code and should be modified for each CPU to be as fast as possible.
+ *
+ * This implementation is 386 version.
+ *
+ *  $Id$
+ */
+
+#undef	ADDCARRY
+#define ADDCARRY(x)     if ((x) > 0xffff) (x) -= 0xffff
+#define REDUCE          {sum = (sum & 0xffff) + (sum >> 16); ADDCARRY(sum);}
+
+/*
+ * Thanks to gcc we don't have to guess
+ * which registers contain sum & w.
+ */
+#define ADD(n)	asm("addl " #n "(%2), %0" : "=r" (sum) : "0" (sum), "r" (w))
+#define ADDC(n)	asm("adcl " #n "(%2), %0" : "=r" (sum) : "0" (sum), "r" (w))
+#define LOAD(n)	asm volatile("movb " #n "(%1), %0" : "=r" (junk) : "r" (w))
+#define MOP	asm("adcl         $0, %0" : "=r" (sum) : "0" (sum))
+
+int
+in_cksum(m, len)
+	register struct mbuf *m;
+	register int len;
+{
+	register u_short *w;
+	register unsigned sum = 0;
+	register int mlen = 0;
+	int byte_swapped = 0;
+	union { char	c[2]; u_short	s; } su;
+
+	for (;m && len; m = m->m_next) {
+		if (m->m_len == 0)
+			continue;
+		w = mtod(m, u_short *);
+		if (mlen == -1) {
+			/*
+			 * The first byte of this mbuf is the continuation
+			 * of a word spanning between this mbuf and the
+			 * last mbuf.
+			 */
+
+			/* su.c[0] is already saved when scanning previous
+			 * mbuf.  sum was REDUCEd when we found mlen == -1
+			 */
+			su.c[1] = *(u_char *)w;
+			sum += su.s;
+			w = (u_short *)((char *)w + 1);
+			mlen = m->m_len - 1;
+			len--;
+		} else
+			mlen = m->m_len;
+		if (len < mlen)
+			mlen = len;
+		len -= mlen;
+		/*
+		 * Force to long boundary so we do longword aligned
+		 * memory operations
+		 */
+		if (3 & (int) w) {
+			REDUCE;
+			if ((1 & (int) w) && (mlen > 0)) {
+				sum <<= 8;
+				su.c[0] = *(char *)w;
+				w = (u_short *)((char *)w + 1);
+				mlen--;
+				byte_swapped = 1;
+			}
+			if ((2 & (int) w) && (mlen >= 2)) {
+				sum += *w++;
+				mlen -= 2;
+			}
+		}
+		/*
+		 * Advance to a 486 cache line boundary.
+		 */
+		if (4 & (int) w && mlen >= 4) {
+			ADD(0);
+			MOP;
+			w += 2;
+			mlen -= 4;
+		}
+		if (8 & (int) w && mlen >= 8) {
+			ADD(0);
+			ADDC(4);
+			MOP;
+			w += 4;
+			mlen -= 8;
+		}
+		/*
+		 * Do as much of the checksum as possible 32 bits at at time.
+		 * In fact, this loop is unrolled to make overhead from
+		 * branches &c small.
+		 */
+		mlen -= 1;
+		while ((mlen -= 32) >= 0) {
+			u_char junk;
+			/*
+			 * Add with carry 16 words and fold in the last
+			 * carry by adding a 0 with carry.
+			 *
+			 * The early ADD(16) and the LOAD(32) are to load
+			 * the next 2 cache lines in advance on 486's.  The
+			 * 486 has a penalty of 2 clock cycles for loading
+			 * a cache line, plus whatever time the external
+			 * memory takes to load the first word(s) addressed.
+			 * These penalties are unavoidable.  Subsequent
+			 * accesses to a cache line being loaded (and to
+			 * other external memory?) are delayed until the
+			 * whole load finishes.  These penalties are mostly
+			 * avoided by not accessing external memory for
+			 * 8 cycles after the ADD(16) and 12 cycles after
+			 * the LOAD(32).  The loop terminates when mlen
+			 * is initially 33 (not 32) to guaranteed that
+			 * the LOAD(32) is within bounds.
+			 */
+			ADD(16);
+			ADDC(0);
+			ADDC(4);
+			ADDC(8);
+			ADDC(12);
+			LOAD(32);
+			ADDC(20);
+			ADDC(24);
+			ADDC(28);
+			MOP;
+			w += 16;
+		}
+		mlen += 32 + 1;
+		if (mlen >= 32) {
+			ADD(16);
+			ADDC(0);
+			ADDC(4);
+			ADDC(8);
+			ADDC(12);
+			ADDC(20);
+			ADDC(24);
+			ADDC(28);
+			MOP;
+			w += 16;
+			mlen -= 32;
+		}
+		if (mlen >= 16) {
+			ADD(0);
+			ADDC(4);
+			ADDC(8);
+			ADDC(12);
+			MOP;
+			w += 8;
+			mlen -= 16;
+		}
+		if (mlen >= 8) {
+			ADD(0);
+			ADDC(4);
+			MOP;
+			w += 4;
+			mlen -= 8;
+		}
+		if (mlen == 0 && byte_swapped == 0)
+			continue;       /* worth 1% maybe ?? */
+		REDUCE;
+		while ((mlen -= 2) >= 0) {
+			sum += *w++;
+		}
+		if (byte_swapped) {
+			sum <<= 8;
+			byte_swapped = 0;
+			if (mlen == -1) {
+				su.c[1] = *(char *)w;
+				sum += su.s;
+				mlen = 0;
+			} else
+				mlen = -1;
+		} else if (mlen == -1)
+			/*
+			 * This mbuf has odd number of bytes.
+			 * There could be a word split betwen
+			 * this mbuf and the next mbuf.
+			 * Save the last byte (to prepend to next mbuf).
+			 */
+			su.c[0] = *(char *)w;
+	}
+
+	if (len)
+		printf("cksum: out of data\n");
+	if (mlen == -1) {
+		/* The last mbuf has odd # of bytes. Follow the
+		   standard (the odd byte is shifted left by 8 bits) */
+		su.c[1] = 0;
+		sum += su.s;
+	}
+	REDUCE;
+	return (~sum & 0xffff);
+}
diff --git a/c/src/exec/libnetworking/netinet/in_cksum_m68k.c b/c/src/exec/libnetworking/netinet/in_cksum_m68k.c
new file mode 100644
index 0000000000..a2889f878a
--- /dev/null
+++ b/c/src/exec/libnetworking/netinet/in_cksum_m68k.c
@@ -0,0 +1,311 @@
+/*
+ * Copyright (c) 1988, 1992, 1993
+ *	The Regents of the University of California.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *	This product includes software developed by the University of
+ *	California, Berkeley and its contributors.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *	@(#)in_cksum.c	8.1 (Berkeley) 6/10/93
+ * $Id$
+ */
+
+#include <sys/param.h>
+#include <sys/mbuf.h>
+
+#if (defined(__GNUC__) && (defined(__mc68000__) || defined(__m68k__)))
+
+#define REDUCE { sum = (sum & 0xFFFF) + (sum >> 16); if (sum > 0xFFFF) sum -= 0xFFFF; }
+
+/*
+ * Motorola 68k version of Internet Protocol Checksum routine
+ * W. Eric Norum
+ * Saskatchewan Accelerator Laboratory
+ * August, 1998
+ */
+int
+in_cksum(m, len)
+	struct mbuf *m;
+	int len;
+{
+	unsigned short *w;
+	unsigned long sum = 0;
+	int mlen = 0;
+	int byte_swapped = 0;
+	union {
+		char	c[2];
+		u_short	s;
+	} s_util;
+
+	for ( ; m && len ; m = m->m_next) {
+		if (m->m_len == 0)
+			continue;
+		w = mtod(m, u_short *);
+		if (mlen == -1) {
+			/*
+			 * The first byte of this mbuf is the continuation
+			 * of a word spanning between this mbuf and the
+			 * last mbuf.
+			 *
+			 * s_util.c[0] is already saved when scanning previous
+			 * mbuf.
+			 */
+			s_util.c[1] = *(char *)w;
+			sum += s_util.s;
+			w = (u_short *)((char *)w + 1);
+			mlen = m->m_len - 1;
+			len--;
+		} else
+			mlen = m->m_len;
+		if (len < mlen)
+			mlen = len;
+		len -= mlen;
+
+		/*
+		 * Force to longword boundary.
+		 */
+		if (3 & (int)w) {
+			REDUCE;
+			if ((1 & (int) w) && (mlen > 0)) {
+				sum <<= 8;
+				s_util.c[0] = *(u_char *)w;
+				w = (u_short *)((char *)w + 1);
+				mlen--;
+				byte_swapped = 1;
+			}
+			if ((2 & (int) w) && (mlen >= 2)) {
+				sum += *w++;
+				mlen -= 2;
+			}
+		}
+
+		/*
+		 * Sum all the longwords in the buffer.
+		 * See RFC 1071 -- Computing the Internet Checksum.
+		 * It should work for all 68k family members.
+		 */
+		{
+		unsigned long tcnt = mlen, t1;
+		__asm__ volatile (
+		"movel   %2,%3\n\t"
+		"lsrl    #6,%2       | count/64 = # loop traversals\n\t"
+		"andl    #0x3c,%3    | Then find fractions of a chunk\n\t"
+		"negl    %3\n\t      | Each long uses 4 instruction bytes\n\t"
+		"andi    #0xf,%%cc   | Clear X (extended carry flag)\n\t"
+		"jmp     %%pc@(lcsum2_lbl-.-2:b,%3)  | Jump into loop\n"
+		"lcsum1_lbl:     | Begin inner loop...\n\t"
+		"movel   %1@+,%3     |  0: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  1: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  2: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  3: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  4: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  5: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  6: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  7: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  8: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  9: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  A: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  B: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  C: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  D: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  E: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n\t"
+		"movel   %1@+,%3     |  F: Fetch 32-bit word\n\t"
+		"addxl   %3,%0       |    Add word + previous carry\n"
+		"lcsum2_lbl:\n\tdbf    %2,lcsum1_lbl   | (NB- dbra doesn't affect X)\n\t"
+		"movel    %0,%3         | Fold 32 bit sum to 16 bits\n\t"
+		"swap    %3             | (NB- swap doesn't affect X)\n\t"
+		"addxw   %3,%0          |\n\t"
+		"moveq   #0,%3          | Add in last carry\n\t"
+		"addxw   %3,%0          |\n\t"
+		"andl    #0xffff,%0     | Mask to 16-bit sum\n" :
+			"=d" (sum), "=a" (w), "=d" (tcnt) , "=d" (t1) :
+			"0" (sum), "1" (w), "2" (tcnt) :
+			"cc", "memory");
+		}
+		mlen &= 3;
+
+		/*
+		 * Soak up the last 1, 2 or 3 bytes
+		 */
+		while ((mlen -= 2) >= 0)
+			sum += *w++;
+		if (byte_swapped) {
+			REDUCE;
+			sum <<= 8;
+			byte_swapped = 0;
+			if (mlen == -1) {
+				s_util.c[1] = *(char *)w;
+				sum += s_util.s;
+				mlen = 0;
+			} else
+				mlen = -1;
+		} else if (mlen == -1)
+			s_util.c[0] = *(char *)w;
+	}
+	if (len)
+		sum = 0xDEAD;
+	if (mlen == -1) {
+		/* The last mbuf has odd # of bytes. Follow the
+		   standard (the odd byte may be shifted left by 8 bits
+		   or not as determined by endian-ness of the machine) */
+		s_util.c[1] = 0;
+		sum += s_util.s;
+	}
+	REDUCE;
+	return (~sum & 0xffff);
+}
+
+
+#else
+/*
+ * Checksum routine for Internet Protocol family headers (Portable Version).
+ *
+ * This routine is very heavily used in the network
+ * code and should be modified for each CPU to be as fast as possible.
+ */
+
+#define ADDCARRY(x)  (x > 65535 ? x -= 65535 : x)
+#define REDUCE \
+  {l_util.l = sum; sum = l_util.s[0] + l_util.s[1];  ADDCARRY(sum);}
+
+int
+in_cksum(m, len)
+	register struct mbuf *m;
+	register int len;
+{
+	register u_short *w;
+	register int sum = 0;
+	register int mlen = 0;
+	int byte_swapped = 0;
+
+	union {
+		char	c[2];
+		u_short	s;
+	} s_util;
+	union {
+		u_short s[2];
+		long	l;
+	} l_util;
+
+	for (;m && len; m = m->m_next) {
+		if (m->m_len == 0)
+			continue;
+		w = mtod(m, u_short *);
+		if (mlen == -1) {
+			/*
+			 * The first byte of this mbuf is the continuation
+			 * of a word spanning between this mbuf and the
+			 * last mbuf.
+			 *
+			 * s_util.c[0] is already saved when scanning previous
+			 * mbuf.
+			 */
+			s_util.c[1] = *(char *)w;
+			sum += s_util.s;
+			w = (u_short *)((char *)w + 1);
+			mlen = m->m_len - 1;
+			len--;
+		} else
+			mlen = m->m_len;
+		if (len < mlen)
+			mlen = len;
+		len -= mlen;
+		/*
+		 * Force to even boundary.
+		 */
+		if ((1 & (int) w) && (mlen > 0)) {
+			REDUCE;
+			sum <<= 8;
+			s_util.c[0] = *(u_char *)w;
+			w = (u_short *)((char *)w + 1);
+			mlen--;
+			byte_swapped = 1;
+		}
+		/*
+		 * Unroll the loop to make overhead from
+		 * branches &c small.
+		 */
+		while ((mlen -= 32) >= 0) {
+			sum += w[0]; sum += w[1]; sum += w[2]; sum += w[3];
+			sum += w[4]; sum += w[5]; sum += w[6]; sum += w[7];
+			sum += w[8]; sum += w[9]; sum += w[10]; sum += w[11];
+			sum += w[12]; sum += w[13]; sum += w[14]; sum += w[15];
+			w += 16;
+		}
+		mlen += 32;
+		while ((mlen -= 8) >= 0) {
+			sum += w[0]; sum += w[1]; sum += w[2]; sum += w[3];
+			w += 4;
+		}
+		mlen += 8;
+		if (mlen == 0 && byte_swapped == 0)
+			continue;
+		REDUCE;
+		while ((mlen -= 2) >= 0) {
+			sum += *w++;
+		}
+		if (byte_swapped) {
+			REDUCE;
+			sum <<= 8;
+			byte_swapped = 0;
+			if (mlen == -1) {
+				s_util.c[1] = *(char *)w;
+				sum += s_util.s;
+				mlen = 0;
+			} else
+				mlen = -1;
+		} else if (mlen == -1)
+			s_util.c[0] = *(char *)w;
+	}
+	if (len)
+		printf("cksum: out of data\n");
+	if (mlen == -1) {
+		/* The last mbuf has odd # of bytes. Follow the
+		   standard (the odd byte may be shifted left by 8 bits
+		   or not as determined by endian-ness of the machine) */
+		s_util.c[1] = 0;
+		sum += s_util.s;
+	}
+	REDUCE;
+	return (~sum & 0xffff);
+}
+#endif