> To: port-hp300@netbsd.org
> Subject: zero-divide during ffs_mountroot()
> From: Jason Thorpe <thorpej@cs.orst.edu>
> Date: Mon, 05 Sep 94 11:20:19 PDT
> Sender: owner-port-hp300@NetBSD.ORG
> 
> ...
> NetBSD 1.0_BETA (HELIX) #1: Mon Sep  5 02:51:55 PDT 1994
>     thorpej@santiam:/tmp_mnt/hurricane/NetBSD/src/sys/arch/hp300/compile/HELIX
> HP9000/433 (33MHz MC68040 CPU+MMU+FPU, 4k on-chip physical I/D caches)
> ...
> scsi0: 32 bit dma, async, scsi id 7
> scsi0 at sc14, ipl 4
> sd9: TOSHIBA CD-ROM DRIVE:XM rev 3332
> sd9 at scsi0, slave 0
> sd0: HP 2213A rev C023, 1296512 512 byte blocks
> sd0 at scsi0, slave 5
> sd1: HP 2213A rev C023, 1296512 512 byte blocks
> sd1 at scsi0, slave 6
> ...
> WARNING: using device at unit 0 of controller
> Changing root device to sd9a
> sd9: scsi sense class 7, code 2, key 2
> sd9: WARNING: unformatted/missing media, defining `c' partition as entire disk
> trap type 3, code = 0, v = 0
> kernel: Zero divide trap
> stopped at   ___qdivrem+0x2c:       movl   d1,-0x22(a6)
> 

Finally took a look at this again.  Someone has changed the interpretation
of "bootdev" in the standalone code but not the kernel.  The original
fragment of stand/autoconf.c:msustobdev() is:

        bdev |= ((rp->msus >> 16) & 0xFF) << B_UNITSHIFT;
        bdev |= (rp->msus & 0xFF) << B_CONTROLLERSHIFT;
        bdev |= (int)hw->hw_pa << B_ADAPTORSHIFT;

while in netbsd it is:

        bdev |= ((rp->msus >> 16) & 0xFF) << B_ADAPTORSHIFT;
        bdev |= (rp->msus & 0xFF) << B_UNITSHIFT;
        bdev |= (int)hw->hw_pa << B_CONTROLLERSHIFT;

in other words, what used to be considered the "controller" (slave 6) is
now considered the "unit" and hence the problems in the kernel's autoconf.

While there is a certain appeal to mapping the SCSI/HPIB slave to the "unit"
rather than the "controller", it is technically wrong.  If any slave were
to support more than one LUN, it would not work.