Pings seem to go to loopback

Forum: Networking
Topic: Pings seem to go to loopback
started by: AW~Flyer

Posted by AW~Flyer on Mar. 01 2006,22:09

I'm stumped by this one.

I picked up some old embedded flat screen PCs lately and figured my best option would be DSL. Pretty much everything works except networking (booting off a frugal install on an IDE compact flash card).

The ethernet chip (cs8900) is not picked up at bootup because it can't be autoprobed. However, "sudo modprobe cs89x0 io=0x300" gets the module installed and I get an eth0 with an occasional "NETDEV WATCHDOG: eth0: transmit timed out" in dmesg.

Netcardconfig from the control panel times out with dhcp. I set sane values by hand but still no joy.

Here's the crazy part. If I try to ping my router it just times out. But on running ifconfig the Rx and Tx packets count for eth0 are 0 and the Rx and Tx packets and byte counts for "lo" have gone up! If I take eth0 down, then it reports network unreachable, so it obviously knows it should be talking to eth0 (the settings in "route" looks sane to my untrained eye).

Help, I'm afraid I'm running out of ideas and things to try.

Posted by cbagger01 on Mar. 02 2006,17:24

If your routing information is incorrect, the packets will not be sent correctly, IE: the traffic will not be sent to the correct network inferface.

If I were you and I still wanted manual networking configuration, I'd use online help guides to better train your eye on the use of "route".

But then again, this could be just a side distraction.

If your router supports DHCP and your netcardconfig DHCP requests time out, then it means that something else is wrong with the picture.

Either:

1) Your hardware is bad.

2) Your connections are bad. Assuming a twisted-pair LAN cable, are the little wires inside the socket making good contact with the LAN cable's plug?

3) You are using the wrong kind of LAN cable. Does your router or network device require a normal "straight-through" LAN cable, or does it require a "crossover" type cable? If your router's ports are described as "auto-sensing" ports, then it doesn't matter because the router will automatically adjust to either type of cable.

4) MOST LIKELY CAUSE: Your network interface driver is not appropriate for this physical card OR it is not configured correctly. You may need to modprobe a different driver module, or you may need to specify additional information besides io=0x300 at the modprobe line. Also, it is possible that the io=0x300 could be the wrong value.

Good Luck.

Posted by cbagger01 on Mar. 02 2006,17:32

FYI,

You also might have an IRQ conflict, see:

< http://www.damnsmalllinux.org/talk/node/274 >

If your network card is connected via a PCI bus, you may be able to get more information by typing:

lspci -v

or maybe by the System Stats application.

Posted by AW~Flyer on Mar. 02 2006,22:55

Thanks for the advice.

I'm still confused why the Rx and Tx counts increase for "lo" not "eth0".

1. 2. 3. Hardware problem?

This was my first reaction. Tried two different, known good, network cables and going straight into router or into 10/100 switch connected to router. I doubt the design is perverse enough to require a crossover cable, though that may be what I try next. The light on the switch for that port lights, which is encouraging, though the device does not appear in the router's arp cache, which is discouraging.

4. The driver is definitely the right one for the chip and 0x300 is the correct I/O location according to the manual, unfortunately it's an ISA bus device and not plug and pray. According to "dmesg" the driver module is seeing the hardware okay and reading the MAC address from the built in eeprom. Thanks for the suggestion of needing more parameters for modprobe: "debug=6" provides more info and I'm working my way through the driver source to find out what it all means.

However, your comment about the interrupt got me thinking. The driver has a lookup table in it which matches the ISA bus interrupt to the interrupt output pin of the CS8900 (there are four interrupt output pins on it) if the connections to the interrupt controller does not match the lookup table (which were taken from the Cirrus Logic reference design) then, even though the IRQ input is correctly selected, the wrong output pin could be generating interrupts. I might give the other three possibilites a try, or prod around the board and buzz out the IRQ lines first. Unfortunately the manual does not include a schematic or list which CS8900 interrupt connects to which IRQ line.

Posted by AW~Flyer on Mar. 03 2006,23:10

Looks like I wasn't doing anything wrong. I have three of these boxes (more or less the same, only the one was set up to boot from compact flash). Swapped the bootable flash to one of the others and that's where I'm typing this from, looks like a hardware problem on the other box.