On Thu, Mar 10, 2005 at 06:29:36PM +0100, Manuel Bouyer wrote:
> > I'll try building from a local disk, my interface is rtk(4) in the client, and ex(4)
> > in the NFS server.
> 
> My second test box also uses rtk(4):
> rtk0 at pci2 dev 5 function 0: Realtek 8139 10/100BaseTX
> rtk0: interrupting at physical irq 11, irq 3
> rtk0: Ethernet address 00:11:2f:3c:87:61
> ukphy0 at rtk0 phy 7: Generic IEEE 802.3u media interface
> 
> I'm going to try some NFS work now.

OK, I didn't get data corruption, but I got several panics in mbuf code;
it looks like mbuf structures are corrupted. I've also seen this without
NFS activity, although NFS seems to be a good way to reliably trigger it.

It could be that the NIC is corrupting data (using wrong addresses on the DMA
descriptors), but I'm suspecting an interrupt problem, which would cause
splfoo() to not block all interrupts. 'systat vm' output doesn't make sense
when the system is busy: lots of timer and softnet interrupts (more than
what there really should be), and some hardware interrupts are not accounted.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--