I've just noticed that the data abort handler fixups for early and late 
abort models will panic if given an instruction with write-back that 
updates, for example, the PC, or is some other UNPREDICTABLE instruction.  
This is bad news, since it means that a user can panic the machine simply 
by executing an unpredictable instruction in a way that will fault.  We 
should raise SIGILL instead of panicing here.

R.