// Copyright (c) 2004-2009 Nokia Corporation and/or its subsidiary(-ies).
// All rights reserved.
// This component and the accompanying materials are made available
// under the terms of "Eclipse Public License v1.0"
// which accompanies this distribution, and is available
// at the URL "http://www.eclipse.org/legal/epl-v10.html".
//
// Initial Contributors:
// Nokia Corporation - initial contribution.
//
// Contributors:
//
// Description:
// in_chk.cpp - IPv6/IPv4 checksum module
//
#include <in_sock.h> // IPv6 enhanced in_sock.h
#include "in_chk.h"
EXPORT_C void TChecksum::Add(RMBufChain &aPacket, TInt aOffset)
{
TMBufIter m(aPacket);
const RMBuf *mbuf;
TUint32 count = 0;
while ((mbuf = m++) != NULL)
{
const TUint16 *p, *end;
const TUint8 *ptr = mbuf->Ptr(), *boundary;
TInt len = mbuf->Length();
union { TUint16 i16; TUint8 i8[2]; } u;
if (len <= 0)
continue;
if (aOffset > 0)
{
aOffset -= len;
if (aOffset >= 0)
continue;
//
// Setup for computing the checksum, starting
// from middle of the buffer. aOffset is left
// with negative value.
//
ptr += (aOffset + len);
len = -aOffset;
}
//
// Collect any solitary leading and trailing octets
// into a single half word and add that first.
//
u.i16 = 0;
// Leading octet
if ((TUint32)ptr & 1)
{
u.i8[1] = *ptr++;
count++;
len--;
}
// Trailing octet
boundary = ptr + len;
if (len & 1)
u.i8[0] = *--boundary;
end = (TUint16*)boundary;
//
// Explicit loop unrolling for checksum calculation.
//
// Duff's device, while not pretty, is perfectly legal
// C++ as well as being the most efficient way to do it.
// The switch statement is used to enter the do-while
// loop at the correct offset. As a result, we get one
// partial iteration of the loop followed by zero or
// more full iterations. Exacly four iterations are
// required to process a completely filled RMBuf.
//
// The gcc used in the current SDK tool chain doesnt't
// seem to be very good at optimizing ARM code, so we
// try to make things simple for it.
//
register TUint32 sum = u.i16;
const TInt loops = (len + 30) >> 5;
if (loops)
{
p = end - (loops << 4);
switch ((len >> 1) & 15)
{
default:
do { //lint -fallthrough
case 0: sum += p[ 0]; //lint -fallthrough
case 15: sum += p[ 1]; //lint -fallthrough
case 14: sum += p[ 2]; //lint -fallthrough
case 13: sum += p[ 3]; //lint -fallthrough
case 12: sum += p[ 4]; //lint -fallthrough
case 11: sum += p[ 5]; //lint -fallthrough
case 10: sum += p[ 6]; //lint -fallthrough
case 9: sum += p[ 7]; //lint -fallthrough
case 8: sum += p[ 8]; //lint -fallthrough
case 7: sum += p[ 9]; //lint -fallthrough
case 6: sum += p[10]; //lint -fallthrough
case 5: sum += p[11]; //lint -fallthrough
case 4: sum += p[12]; //lint -fallthrough
case 3: sum += p[13]; //lint -fallthrough
case 2: sum += p[14]; //lint -fallthrough
case 1: sum += p[15];
} while ((p += 16) < end);
}
}
//
// Swap byte order if there is an alignment mismatch
// between this buffer and the previous ones. We fold
// the partial sum into the lower half word in order
// to avoid the possibility of overflow.
//
if (count & 1)
{
iSum += (sum >> 24); // add octet 3
iSum += (sum >> 8) & 0xffff, // add octets 2-1
iSum += (sum << 8) & 0xff00; // add octet 0
}
else
{
iSum += sum;
}
// Update number of bytes summed
count += len;
}
}
EXPORT_C void TChecksum::Add(RMBufChain &aPacket, const RMBufPktInfo *aInfo, TInt aOffset)
{
if (aInfo)
{
//
// Compute the pseudo header part
//
// It is assumed that the aInfo->iLength is the length
// of the full IPv6 packet, and the aOffset gives the
// beginning of the upper layer header. For the pseudo
// header computation, the upper layer length must be
// used and thus the aOffset is subtracted!
//
AddH((TUint32)aInfo->iLength - aOffset);
AddH((TUint16)aInfo->iProtocol);
Add((TUint16 *)&TInetAddr::Cast(aInfo->iSrcAddr).Ip6Address(), 16);
Add((TUint16 *)&TInetAddr::Cast(aInfo->iDstAddr).Ip6Address(), 16);
}
//
// Add the payload part
//
Add(aPacket, aOffset);
}
EXPORT_C TUint32 TChecksum::Calculate(const TUint16 *aPtr, TInt aLen)
{
register TUint32 sum = 0;
if (aLen >>= 1)
{
//
// Explicit loop unrolling for checksum calculation.
//
// Duff's device, while not pretty, is perfectly legal
// C++ as well as being the most efficient way to do it.
// The switch statement is used to enter the do-while
// loop at the correct offset. As a result, we get one
// partial iteration of the loop followed by zero or
// more full iterations. Exacly four iterations are
// required to process a completely filled RMBuf.
//
// The gcc used in the current SDK tool chains doesnt't
// seem to be very good at optimizing ARM code, so we
// try to make things simple for it.
//
const TUint16 *end = aPtr + aLen;
const TInt loops = (aLen + 15) >> 4;
register const TUint16 *p = end - (loops << 4);
switch (aLen & 15)
{
default:
do { //lint -fallthrough
case 0: sum += p[ 0]; //lint -fallthrough
case 15: sum += p[ 1]; //lint -fallthrough
case 14: sum += p[ 2]; //lint -fallthrough
case 13: sum += p[ 3]; //lint -fallthrough
case 12: sum += p[ 4]; //lint -fallthrough
case 11: sum += p[ 5]; //lint -fallthrough
case 10: sum += p[ 6]; //lint -fallthrough
case 9: sum += p[ 7]; //lint -fallthrough
case 8: sum += p[ 8]; //lint -fallthrough
case 7: sum += p[ 9]; //lint -fallthrough
case 6: sum += p[10]; //lint -fallthrough
case 5: sum += p[11]; //lint -fallthrough
case 4: sum += p[12]; //lint -fallthrough
case 3: sum += p[13]; //lint -fallthrough
case 2: sum += p[14]; //lint -fallthrough
case 1: sum += p[15];
} while ((p += 16) < end);
}
}
return sum;
}