On 13 Jul, Miguel Mendez punched keys in this particular order:
>> 	trap: TLB miss (load or instr. fetch) in kernel mode
>> 	status=0x40003, cause=0x10000000, epc=0x0, vaddr=0x0
>> 	pid=82 cmd=dev_mkdb usp=0x0 ksp=0xca567e88
>> 	Stopped in pid 82 (dev_mkdb) at          0:   invalid address.
>> 	db>
> 
> Tha's the problem I was seeing until today. For some reason or other,
> the 3.2GB Quantom Fireball disk I was using has died, so I've replaced
> it with the 1GB SGI-branded IBM disk that came with the machine, and,
> voila! Now the machine finishes the install without any problem. I'm
> using 1.6.1 on it right now, but will upgrade it to current at some
> point. Perhaps your problem is also hardware related? I'm sure how a
> defective disk could cause a TLB related panic, but it seems to be gone
> now. Thanks to all for all the help.

Well, it won't be the disk, as there isn't one in it. The whole machine
runs from NFS mounted filesystems.
But maybe the problem could be triggered by read/write delays or errors,
when the NFS server is unavailable for a split second or so.

It looks like the TLB misses are related to the virtual memory system
(though I just looked up the word in google). In that case, you
probably had swap problems with that bad harddisk. I don't have swap at
the moment at all, so it could have a similar cause. But I'm
speculating...

I'll try to give it some swap somehow. That way possible kernel core
dumps can also be saved.

-- 

Alban Hertroys                              http://solfertje.student.utwente.nl
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
                  CORBA,    A poisonous snake that bites.