One further comment:
> This patch appears to fix my machine check problem on the 4kc. The 4kc
> shouldn't need an ssnop here, but this appears to fix it.
The 4Kc shouldn't need an ssnop/ehb for the purposes of a *load*, but
if the exception was on an instruction fetch, the documentation is reasonably
clear that there's a hazard of 3 cycles before the new entry can be used for
a fetch - which is what ERET will do in the case of a instruction fetch miss.
Regards,
Kevin K.
