On Tue, 05 Nov 1996 13:08:00 EST,
briggs@puma.macbsd.com rearranged the electrons to say:
>
>> 1)  What does the FPSP compile option do?  040 floating point software?
>>     Do I need this to use an 040 CPU?
>
>FPSP stands for Floating Point Software Package, I think.  It's a library
>(from Motorola) for the 040 to emulate the FPU operations that the '881
>and '882 implement but which were left out of the 040's FPU.

We were getting "illegal instruction" errors after booting into single-user
with the Daystar, when executing commands like 'w' and 'ps'.  Would enabling
FPSP in the kernel possibly fix this problem?

>> 2)  What compile options must be set to boot an 040?
>
>The GENERIC kernel _should_ boot on an 040.  I don't know that it will.

What Tony should have asked is "what options should we enable that
normally are not enabled (for our IIci)", e.g. is FPSP necessary?
This is a IIci and we have been using ncrscsi, would a switch to sbc
be prudent?

The only reason we ask is that it takes a couple hours to compile a
kernel from scratch, which is a wee bit long to just try options on a
whim. :-)

>> 3)  Is there anyway to boot from a different partition of the same
>>     drive?  We set up a partition where we could set up a different
>>     root partition for testing purposes, but we don't know how to
>>     boot it.
>
>Try using different partition names.  The booter should scan for the
>partition name specified in the Booting... preferences and, if it fails
>to find that, go for the first root partition on the drive.

we tried this, but were unsure as to what "partition name" really means,
is it "/dev/sd0e", is it "/tmp"?  (neither appeared to work).

Actually this is really important, because the only external drive
we have right now to test is an old Quantum 240, which for whatever
reason the $%#&^$ disklabel thinks has 65 sectors/track instead of
87.  Most of the time the kernel just bombs when booting with the Daystar
card off the Quantum, on the "mode sense(4) returned fictitous geometry"
error.  Attempts to write a correct disklabel have failed miserably.

>One possible problem with the 040 is that the 030 and 020 have a
>"dynamic bus sizing" feature which allows the processor to, for
>instance, write a long word to the data bus and have a byte-oriented
>device (like the ncr5380) read the 4 bytes in 4 read cycles.  This
>could screw up the current SCSI implementation if a SCSI device
>disconnects on a non-4-byte boundary and if Daystar didn't include
>some kind of emulation for this feature.  Incidentally, I think that
>Apple provides this on some of the later Quadras with the PrimeTime IC.

I think the problem is more related to the "illegal instruction" errors
we've been getting.  I posted the results a while ago, but I'm not
sure if you saw it.

If someone could give us a hint about booting off of our /tmp partition,
we can get back to testing.  Or, is there a way to write a correct
disklabel to the Quantum that the kernel keeps believing has the wrong
geometry?

Sean

__________________________________________________________________________
Sean Sweda                        :            Owner, Motor City Marauders
sweda@us.itd.umich.edu            :   1994 Internet Baseball League Champs
http://www.umich.edu/~sweda       :                     http://www.ibl.org
__________________________________________________________________________