On Sun, 17 Mar 1996 15:04:15 -0800 (PST) 
 Paul Goyette <paul@pgoyette.bdt.com> wrote:

 > >Synopsis:       new config changes fail on mac68k port

 > >Description:
 > 	Get following messages when booting new kernel:
 > 		ARGH!!  No boot device????Only supports DISK device for booting.
 > 		Changing root device to e.
 > 		panic: cannot mount root

After a discussion with Scott Reynolds, I think I know what the problem 
is, and quite frankly, I'm *very* surprised the drivers ever worked in 
the first place!

Short version of the story: the mac68k "mainbus" driver uses "indirect 
autoconfigration" to attach children.  This changes the second argument 
(the "match" argument) to the match/probe routines from a "struct cfdata *"
to a "struct device *" ... the ncr5380 drivers attach to mainbus on a mac.

HOWEVER!  *BOTH* of the 5380 drivers were accessing the "match" argument 
as a "struct cfdata *"!  This, AFAICT, is how they've *always* worked, 
and when they did the "cf->cf_unit == 0" check, they must have gotten 
extremely lucky and just happened to read something that had a value of 0.

I "fixed" the 5380 drivers (read: made them cope with "indirect") and 
sent Scott patches this morning.  I assume that he'll test them tonight 
and check them in if they work.  If anyone else wants these patches in 
the mean time, contact me privately.

BTW, this is a shining example of how the "indirect" hack can screw 
people up, and why we should fix all of the drivers that rely on it and 
nuke if from orbit.

--------------------------------------------------------------------------
Jason R. Thorpe                                       thorpej@nas.nasa.gov
NASA Ames Research Center                               Home: 408.866.1912
NAS: M/S 258-6                                          Work: 415.604.0935
Moffett Field, CA 94035                                Pager: 415.428.6939