Re: Netbooting a sparc64 (netbsd-6)

On Feb 27, 2012, at 21:15 , Chris Ross wrote:
>> There is an easy way to verify the problem I had in mind: connect another
>> port as well and use that to netboot - e.g. have the prom use bge1 to load
>> the kernel, then the kernel will be able to use bge0.
> 
>  The first attempt to do this, with a manual "move the cable between ports, and use the same IP address for both" didn't work.  If I get too much further, I'll make a new cable (it's a long run, this server isn't in the service area/room/closet yet) and try it the hard way.  Would you suggest using two different IP addresses, given that there are two MACs ?

  In case this helps, continuing the "switch the cable and wait, after the kernel fails to mount the nfs filesystem".

  I loaded the bootloader via tftp, and the kernel via NFS, over bge1.  Then, after the kernel loads and fails to configure bge1, I switched the cable to bge0.  I waited a while, checked that the arp cache on the router didn't have anything, and cleared it on the NFS (/rarpd/tftpd) server.  I told the sparc to use bge0, and it claims to be trying to get an address via DHCP/BOOTP, but there is never any traffic from the mac(s) on that machine [seen by the router/DHCP server].

  So, it looks more like this version of the kernel (netbsd-6, circa 25-Feb) just isn't passing packets at all.  So, with the kernel in question, I loaded the 5.99.59 userland I had on a disk.  I then tried to dhcpcd from it, and also saw nothing on the dhcp server.

  So.  I then cold-booted into the 5.99.59 image on the disk, which was never really fully configured.  It's in full Amnesiac mode.  If I run dhcpcd on bge0, it gives up assigning a 169.254/16 address, my dhcpd never having seen a packet.  But, if I kill dhcpcd and manually configure an address, it seems to pass packets, on net and through the router.

  This is smelling more like a sparc64 bge bug as time goes on...

  Anyone have any new thoughts or suggestions?

              - Chris