[Gumstix-users] eth0 breakdown when used parallel to wlan0? (smc911x issue?)

[Gerhard H. <Ger...@pe...>]

Hi all,

we are working on an application which receives a data stream with up to
700kbit/s from eth0, buffers it to a sd card and sends it again via
wlan0. Receiving and buffering the data needs to run in realtime.

At the moment we are using verdex pro XL6 together with the netpro-vx.
The application is written in java running at jamvm1.5.0 together with
classpath 0.96.1. Generally we are using a self build rootfs which
includes the jamvm and classpath additionally to the basic image.

We were developing and testing for weeks and the eth0-wlan0 link was
working quite fine. Very seldom we got the problem, that eth0 stoped
working - both link LED's are off. The only way we get to run eth0 again
in this situation is removing and adding the smc911x module using
"modprobe -r smc911x" and "modprobe smc911x"...
But since last week it stops working after around 1 Minute

... This all feels quite similar than the problem described by Paul in
April (see thread
http://www.nabble.com/Verdex-board-smc911x-issue-td23142184.html), but
it isn't in detail, because I already removed the SMC91x Kernel modules
and I do not get this startup warnings ...
-------------
gumstix_smc91x_init: smc91x chip not found, returning -ENXIO
smc91x: Unknown symbol gumstix_smc91x_load
eth0: LAN9117 (rev 2) at 0xc8870000 IRQ 131
eth0: Ethernet addr: 00:15:c9:0e:c8:50
-------------
... any more, but I still have the eth0 breakdown.

It started to get worst and worst. At the beginning I had only problems
when eth0 was connected to an 10MBit halfduplex device, where it stopped
working after around 5 Minutes. Connected to a 100MBit/s Ethernet Device
I had no problems.... spooling with 600kBit/s (I stopped the tests after
2h most of the time). Now it stops already after 1 Min even on 100MBits
eth devices
A few days ago I got quite often the kernel message "|NETDEV WATCHDOG:
eth0: transmit timed out", but not any more. Now it dies silent.|

I already tried
1) to move the buffer from the sd card to the RAM disk (because maybe
the sdcard support has problems when doing parallel read/write)
2) reflashing the factory image
Angstrom-gumstix-basic-image-glibc-ipk-2007.9-test-20080512-gumstix-custom-verdex.rootfs.jffs2
from gumstix.net and installing jamvm and classpath packages on the sd card
3) activate ip-forwarding from eth0 to wlan0 by the kernel only (no java
application was started at all)

All that didn't change the eth0 breakdown behaviour after around 1 Min.

Does anybody have an idea what could go wrong here?

cheers
Gerhard