/*-
* Copyright (c) 1990 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
* 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.
*
* from tahoe: in_cksum.c 1.2 86/01/05
* from: @(#)in_cksum.c 1.3 (Berkeley) 1/19/91
*/
#ifndef VBOX
__FBSDID("$FreeBSD: src/sys/i386/i386/in_cksum.c,v 1.28.10.1.6.1 2009/04/15 03:14:26 kensmith Exp $");
/*
* MPsafe: alfred
*/
#include <netinet/in_systm.h>
#include <machine/in_cksum.h>
#else
# include "in_cksum.h"
# include "slirp.h"
#endif
/*
* 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.
*/
/*
* icc needs to be special cased here, as the asm code below results
* in broken code if compiled with icc.
*/
#if !defined(__GNUCLIKE_ASM) || defined(__INTEL_COMPILER)
/* non gcc parts stolen from sys/alpha/alpha/in_cksum.c */
#define REDUCE32 \
{ \
}
#define REDUCE16 \
{ \
}
#endif
#if !defined(__GNUCLIKE_ASM) || defined(__INTEL_COMPILER)
/*0 bytes*/ /*1 byte*/ /*2 bytes*/ /*3 bytes*/
0x00000000, 0x000000FF, 0x0000FFFF, 0x00FFFFFF, /* offset 0 */
0x00000000, 0x0000FF00, 0x00FFFF00, 0xFFFFFF00, /* offset 1 */
0x00000000, 0x00FF0000, 0xFFFF0000, 0xFFFF0000, /* offset 2 */
0x00000000, 0xFF000000, 0xFF000000, 0xFF000000, /* offset 3 */
};
union l_util {
u_int16_t s[2];
u_int32_t l;
};
union q_util {
u_int16_t s[4];
u_int32_t l[2];
u_int64_t q;
};
static u_int64_t
{
int offset;
return sum;
}
if (len <= 0) {
return sum;
}
}
#if 0
/*
* Force to cache line boundary.
*/
if (4 & offset) {
lw += 1;
}
if (8 & offset) {
lw += 2;
}
if (16 & offset) {
lw += 4;
}
}
#endif
/*
* access prefilling to start load of next cache line.
* then add current cache line
* save result of prefilling for loop iteration.
*/
lw += 8;
}
if (len >= 0) {
lw += 8;
} else {
len += 32;
}
while ((len -= 16) >= 0) {
lw += 4;
}
len += 16;
while ((len -= 4) >= 0) {
}
len += 4;
if (len > 0)
return sum;
}
{
return (sum);
}
{
return (sum);
}
{
int mlen = 0;
int clen = 0;
goto skip_start;
} else {
}
}
if (m->m_len == 0)
continue;
else
}
return (~sum & 0xffff);
}
{
return (~sum & 0xffff);
}
#else
/*
* These asm statements require __volatile because they pass information
* via the condition codes. GCC does not currently provide a way to specify
* the condition codes as an input or output operand.
*
* The LOAD macro below is effectively a prefetch into cache. GCC will
* load the value into a register but will not use it. Since modern CPUs
* reorder operations, this will generally take place in parallel with
* other calculations.
*/
struct mbuf *m;
int len;
int skip;
{
register u_short *w;
register unsigned sum = 0;
register int mlen = 0;
int byte_swapped = 0;
goto skip_start;
} else {
}
}
if (m->m_len == 0)
continue;
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
*/
w = (u_short *)((char *)w + 1);
len--;
} else
/*
* Force to long boundary so we do longword aligned
* memory operations
*/
if (3 & (int) w) {
if ((1 & (int) w) && (mlen > 0)) {
sum <<= 8;
su.c[0] = *(char *)w;
w = (u_short *)((char *)w + 1);
mlen--;
byte_swapped = 1;
}
sum += *w++;
mlen -= 2;
}
}
/*
* Advance to a 486 cache line boundary.
*/
"addl %1, %0\n"
"adcl $0, %0"
: "+r" (sum)
: "g" (((const u_int32_t *)w)[0])
);
w += 2;
mlen -= 4;
}
"addl %1, %0\n"
"adcl %2, %0\n"
"adcl $0, %0"
: "+r" (sum)
: "g" (((const u_int32_t *)w)[0]),
);
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) {
/*
* 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.
*/
"addl %1, %0\n"
"adcl %2, %0\n"
"adcl %3, %0\n"
"adcl %4, %0\n"
"adcl %5, %0\n"
"mov %6, %%eax\n"
"adcl %7, %0\n"
"adcl %8, %0\n"
"adcl %9, %0\n"
"adcl $0, %0"
: "+r" (sum)
"g" (((const u_int32_t *)w)[0]),
: "eax"
);
w += 16;
}
if (mlen >= 32) {
"addl %1, %0\n"
"adcl %2, %0\n"
"adcl %3, %0\n"
"adcl %4, %0\n"
"adcl %5, %0\n"
"adcl %6, %0\n"
"adcl %7, %0\n"
"adcl %8, %0\n"
"adcl $0, %0"
: "+r" (sum)
"g" (((const u_int32_t *)w)[0]),
);
w += 16;
mlen -= 32;
}
if (mlen >= 16) {
"addl %1, %0\n"
"adcl %2, %0\n"
"adcl %3, %0\n"
"adcl %4, %0\n"
"adcl $0, %0"
: "+r" (sum)
: "g" (((const u_int32_t *)w)[0]),
);
w += 8;
mlen -= 16;
}
if (mlen >= 8) {
"addl %1, %0\n"
"adcl %2, %0\n"
"adcl $0, %0"
: "+r" (sum)
: "g" (((const u_int32_t *)w)[0]),
);
w += 4;
mlen -= 8;
}
if (mlen == 0 && byte_swapped == 0)
continue; /* worth 1% maybe ?? */
while ((mlen -= 2) >= 0) {
sum += *w++;
}
if (byte_swapped) {
sum <<= 8;
byte_swapped = 0;
if (mlen == -1) {
su.c[1] = *(char *)w;
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)
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;
}
return (~sum & 0xffff);
}
#endif