Reenter KDB on fault on powerpc, instead of panicking

Most other architectures already re-enter KDB on faults, powerpc and mips
are the only outliers. Correct this for powerpc, so that now bad addresses
can be handled gracefully instead of panicking.