On Thu, May 10, 2001 at 09:04:03PM -0000, eeh@netbsd.org wrote:
	> 
	> 	I got my DEBUG enabled kernel crashed with the attached test,
	> 	first it breaks system to ddb and an attempt to continue
	> 	ends up in ofw 'ok' prompt. Commenting out lines below helps.
	> 	locore.s:4795
	> 	#ifdef DEBUG
	> 		tst	%g2
	> 		tz	1		! tpc NULL? Panic
	> 		tst	%i6
	> 		tz	1		! %fp NULL? Panic
	> 	#endif
	> 
	> That check is there to make sure the cpu is not trying to return from a
	> trap with either a NULL pc or stack pointer.  The machine is unlikely to
	> recover in that event, and this the trap lets you determine what the problem
	> is.
	> 
	That's what I started with. If I change DEBUG to DEBUG_NOTDEF
	then the same test doesn't cause any of the problems, i.e even
	%g2 is still 0 it doesn't lead to crash.

	I concluded that %g2 at this point not necessarily will become tpc
	and that trap to debugger won't return normally.

That's interesting.  As far as I can tell, those values will eventually
be loaded into those registers.  If not immediately, then they will be
stored in a location that will eventually be loaded into the stack pointer
or tpc register.

Anyway, I would be interested to know which register is zero and how it
got to be set to zero.  It would be better to fix the root cause than
work around a potential bug.

Eduardo