Linux NFS ate my bge

[Chris Hedley]

My bge network interface dies when faced with a barrage of Linux NFS 
requests.

I've (re)discovered this now that I've recently got my assorted config 
problems largely ironed out and have my FreeBSD box up to date again, I'm 
reminded of an unresolved problem from way back, which is that my bge card 
collapses after being subjected to a large amount of NFS traffic coming 
from my Linux box, e.g. recompiling KDE on a discless workstation, which 
has been responsible for three embuggerances so far today.

Now I'm not sure if it's specifically Linux NFS it doesn't like, or just 
NFS, or even just a lot of traffic (though Samba doesn't do this - I'd 
like to say it's probably because Windows' CIFS or whatever it's called 
this week is too inefficient to generate that amount of traffic, but 
actually its performance is much better than FreeBSD - Linux NFS), but it 
frequently puts my FreeBSD system into a state where my bge is unusable 
unless I reboot; and by "reboot" I mean "reset" as it seems to render the 
rest of the system prone to locking up solid, presumably putting anything 
wanting network access into a no-wake state, or whatever the 21st century 
name is for an unkillable process.

Scanning back through the archives I see I'm not alone with both this 
problem and assorted other bge-related annoyances, which I gather are 
likely due to insufficient documentation being provided to the kernel devs 
to get a robust driver together.

I guess my question is "have any breakthroughs been made?" and given that 
I'm using -current, I guess the answer is "no".

In which case I guess my next question is "what cheap but good PCI/PCI-X 
(but not PCIe) based card would you recommend?" since I'd rather just get 
this one sorted out once and for all.  My only real requirement is that it 
does 1 Gbit.  And works.




I did wonder for a while if it might be something to do with having four 
aliases on the same subnet on my bge (32 bit netmask, of course) but it 
seems to make no difference whether they're configured or not (it was 
interesting seeing the ifconfig after one bge failure place the primary 
interface /after/ all its aliases; don't know if that's symptomatic of 
what's going on, but do bear in mind that as I mentioned it seems to make 
no difference to its stability whether those aliases are present or not).




Another (totally unrelated) thing that should probably go in its own 
thread, I've found an oddity with the ohci and/or ukbd drivers which is 
that if I compile them into my kernel it won't boot, but if I leave them 
as separate modules and get loader to deal with them it boots fine. 
That's "won't boot" as in "gets to the orm0 bit of its boot messages and 
then hangs"; normally sc0 comes next, though I don't know if the USB bits 
change that.  I'm not that bothered as I prefer modular kernels anyway, 
but I thought I'd bring it up in case anyone's interested.

Cheers,

Chris.