Coldfire MCF5485EVB eval board with Linux 2.6.10 Kernel and SNMP problem

[jkimble]

I'm working with NET-SNMP trying to get a working compile for this platform. Has anyone managed to get SNMP to work on Coldfire (MCF5485) running Linux (2.6.10 kernel). If so, what version of SNMP were you using and what compile flags were used to configure it? What I'm doing seems like it should work (it compiles) but nothing will run (bus error).

Any help much appreciated....

[truregs]

We're trying to get net-snmp(d)  runing on Coldifre (5484) with a linux 2.6.10 from the 2007 BSP.

We're using 5.2.5.1 net-snmp release.

our configuration flags are : 

--prefix=$DEMSV_DIR/bin/$DEMSV_SYSTEME/net-snmp \
--host=m68k-linux-gnu \
--with-endianness=big \
--with-cflags="-g -O0" \
--enable-shared=no \
--enable-static=yes \
--enable-debugging=no \
--with-default-snmp-version="2" \
--enable-silent-libtool \
--with-libs="$LIB_OC_SHM" \
--includedir="$DEMSV_DIR/INT/H/" \
--with-mib-modules=" \
our mib module list..." \
--with-logfile="/tmp/snmp.log" \
--with-persistent-directory="/tmp" \
--with-copy-persistent-files="yes" \
--with-perl-modules="PREFIX=$DEMSV_DIR/bin/$DEMSV_SYSTEME/net-snmp/lib/perl-snmp"

If we use the default kernel paging options ( 8 pages locked in m68k/coldfire/head.S ), 

we experienced quite good performances, but segv and bus error happens quite fast, depending on the mib-oid being interrogated.

 

Raising this parameter to 16 pages results in response time increase by a factor of 5, but the bug disappears.

Using gdb on target, points some problems in various functions  : strstr, free, ... thus not allways related directly to memory operation.

Do you have any clue ?