>>But "machine check"s are _not_ the same as "correctable error"s...
>
> Fair enough ... but in this case, they seem to be related .... after
> I rack up a bunch of correctable errors, then I get a machine panic ...
> but if I'm doing stuff that doesn't trigger correctable errors, I'm
> perfectly fine.

Are you saying you can cause correctable errors on demand?  How?

If these are single-bit memory errors being corrected by ECC,
you likely have a DRAM going bad or possibly a bad connection.
If possible, it is useful to have the location logged so that
you can reseat/replace the bad memory chip/SIMM/board/whatever before
you get a double-bit error and panic.  This isn't necessarily something
you want output to the console, unless that's the only option.

Perhaps the "machine panic" is a double bit (uncorrectable) error?

-Dave