William Montgomery wrote:
Thanks for responding. I am very interested to find the source of this problem.
Kok, Auke wrote:
William Montgomery wrote:
In an earlier post to the list I described a hard lockup condition
that occurs on linux kernels 2.4.22, 2.6.13, and 2.6.17 when using
a 4 port 10/100 fast ethernet card. The lockup is easily repeatable
and occurs on 2 out of 3 computers.
Further testing has revealed that the lockup can be prevented on all
computers by making sure the card is installed on the primary PCI bus.
If the card is installed in a slot on the secondary PCI bus (behind a
PCI to PCI bridge) the lockup occurs.
sounds like int-A/B/C/D routing issues
The strange thing is that all the ports on the card work fine for a few minutes, then when some condition (as yet unknown) occurs the system locks up hard. I am currently using a PCI bus analyzer to capture bus activity just prior to the lockup to try and find out what leads up to this condition.
are you running any form of irqbalance, either in-kernel (bad) or the userspace (better) one?
Are there any PCI tuning registers that I can tweak to get around
this problem? Any changes I could make to the e100 driver to fix this?
this issue might be resolvable by quirking the bridgee chips and adjusting any APIC where needed. Unfortunately I don't know much about this but it's physically not possible from the e100 driver. The special (non-intel) card that has these 4 ports onboard contains a bridge chip itself which explains the issues. Even a BIOS issue could be the cause here.
I am aware of the bridge chip on the card but not sure what you mean when you say this explains the issues? I sure would like to figure out a way around this.
irq routing in linux may not be the same as in windows. I have no idea how to compare them either (dmesg will show the linux setup, but I don't know how to retreive this info under linux).