Troubles with adaptec 62044 card and starfire driver (something wicked happened)
David Jorrin
fast-Ethernet card (aka DuraLan).
I'm suffering two anomalous situations:
- Sometimes the driver reports "something wicked happened" errors. The
error code always is 204xxxx.
- Eventually the driver reports "transmit timed out" errors and the
NIC hangs. It works again after an ifconfig's down and up. I read the
driver's timeout code and it should reset the nic but actually
disables it.
Donald Becker explains me that the 204xxxx error reports a normal
driver's behaviour for avoiding a Tx FIFO underrun. Also he pointed me
about some IRQ problem related with APIC on SMP systems.
So, I tried the next solutions:
- I recompiled kernel with SMP support because I'm using a dual-cpu
motherboard with 1 processor (HP NetServer LPr). The new kernel uses
APIC IRQ management.
- Furthermore I used another network card with different IRQ's that
isn't shared with other devices.
However the problem remains without any visible change. Since the new
nic now has its own IRQ, I think that is not an interrupt problem.
Also I tried to test if could be a performance problem with a flood
ping (ping -f -s 2048) from the Linux box. I get these
results:
- The flood ping against the router always produces a few "something
wicked" errors. Also it increases the nic counters for carrier and
collisions. However the ping reports a 0% lost packets and the nic
doesn't hang.
- The same flood ping against another host (with a lower traffic)
doesn't produce any error.
- Once, the flood ping (against the router) produces the "transmit
timed out" error and hanged the nic.
I have read other posts about transmit timed outs but related with
other drivers. The solution was a driver patch or configure the kernel
with the "CPU is too slow to handle bandwith" option. It's no useful
in this case because I'm using the last starfire driver's version
(v1.03) and a Pentium III 550 MHz.
I wonder if the problem could be produced by:
- Some error in th network switch or vlan (congestion, collisions, etc.)
- Or some bug in starfire driver related with high transfer conditions.
I would appreciate very much all your suggestions to detect the problem.
Best regards,
David Jorrin.
In article <8pbflq$fmp$1...@nnrp1.deja.com>,
David Jorrin <David....@writeme.com> wrote:
> Dear linux kernel gurus,
>
> I have a problem with my linux box and the Adaptec 62044 four-port
> fast-Ethernet card (aka DuraLan):
>
> - I installed a RH 6.2 with kernel 2.2.16-3 and the last Donald Becker
> starfire driver (starfire.c:v1.03 7/26/2000) compiled as kernel
> module.
>
> - The box has 4 Adaptec cards (yes, 16 Ethernet ports) but we have
> configured only 4 interfaces (eth0, eth1, eth2, eth15)
>
> - Also it has a ncr53c8xx scsi controller.
>
> - Every cards and interfaces are properly detected and configured at
> kernel boot. Initially everything seems work right initially.
>
> - After some time, it's detected the kernel error: "eth0: Something
> Wicked happened! xxxxxxx". At those moments, the network link is
> momentarily frozen but apparently it resumes a few seconds after. This
> situation repeats some times with different error codes (2048101,
> 2049001, 2049101, 204a001, 204b001).
>
> - Finally, the eth0 interface is hanged and the kernel reports every
> few seconds the next error: "eth0: Transmit timed out, status
> 00000000, resetting...". Each error message is followed by a long hex
> dump.
>
> I have read the support information for the starfire driver but I
> don't understand the error's cause. Reading some post related with
> the "something wicked" message, I deduce that could be related with
> some IRQ or performance conflict with other device. However I didn't
> find any solution.
>
> I'm really concern about because we need use the 16 interfaces.
>
> I would appreciate any help.
>
> Best regards,
>
> David Jorrín.
>
Sent via Deja.com http://www.deja.com/
Before you buy.