Here are a few things I have discovered since my earlier message.


Allen Briggs at briggs@ninthwonder.com writes:

>> .. Although it seems to crash more reliably with "options NMBCLUSTERS=1024"
> 
> Hmmm...
> 

I take that back! I have had about the same track record with this version
of the kernel. It appears to be quite random as to when it decides to panic,
but always while downloading an mp3. Larger files have been transferred in
other formats and thru different ports with no problem.

>> It is definitely a "scsibus at ncrscsi0" situation. I also forgot to mention
>> that there are 2 NuBus Ethernet cards in the machine. One is a SONIC and I'm
>> not sure of the nature of the other. One has 16K of memory and the other has
>> 32K.
> 
> One's definitely not a SONIC with dp8390 stuff in the stack.  Could you
> run:
> dmesg | grep ae[0-9]
> and    dmesg | grep sn[0-9]
> 
> and send the results to port-mac68k?  Or just send the entire dmesg
> output.  I'd like to know if your 'ae' ethernet interface has 16k...
> 

You're absolutely correct. I must have been thinking about another machine.
There is no Sonic Card here.

isis:/var/crash #dmesg | grep ae[0-9]
ae0 at nubus0 slot d: MacCon NuBus-A , 32KB memory
ae0: Ethernet address .....
ae1 at nubus0 slot e: Ethernet A-Series, 16KB memory
ae1: Ethernet address .....

These are the only NICs in the machine.

>> I also got the output from the "trace" command in db. You mentioned ddb, but
>> after the crash I get a db> prompt. I imagine it's the same thing.
> 
> It is.
> 
>> _dp8390_ring_copy(469400,3c04,2a6802,5ea) + 40
> 
> OK.  This is trying to copy 1514 bytes starting at offset 15364 in the
> card.  If the card has 16K, this might be a problem since this will take
> us 494 bytes past 16K.  That should not be a problem--it looks like the
> code handles that case, but it is suspicious.
> 

Fascinating!

>> I had to copy this using pen and paper, because I saw no realistic way to
>> dump it to a file. I tried to look at the dumped core...
> 
> You don't have a serial console or serial boot echo option?
> 

I do but am completely unfamiliar with ddb as of now. How can you recall a
trace from a previous crash after the machine has rebooted and you break
into the debugger?

>> Does this mean that my crash dump is useless? I expected to be able to get
>> the "trace" output from these core files, or something similar, anyway.
> 

This stuff is extremely interesting to me. I have very little experience
with issues of the kernel. Where could one obtain more detailed information?

Louis