> 	Well.. I have good news and bad news. With ignore_bogus_traps set to
> 	1 it passes that crashme area without problem, but later on the
> 	machine hangs hard (L1+A no effect). 
> 
> 	The necessary magic to kill it this time was:
> 		crashme +2000.0 816 100


I can confirm that a sparcstation 5 can be made to crash this way. I've also
noticed that there is no pattern in the location in the code where the
fault happens or an inconsistency is detected.  It all looks suspiciously
like the register file corruption related to MMU activity (which is
on-chip in a SS5) that has been playing up before.


The sun4 code (crashing IPX that started this thread) has been fixed.
That bug had nothing to do with the above.

-pk