> It would be cool to have an interrupt priority level trace facility
> that logged levels and cycle counter times.  This would pretty clearly
> have to be a special compile of the kernel to replace/adjust the spl()
> calls and catch interrupt-routine entry/exit.

Actually it is probably enough to sort out the return address
of the serial interrupt that reports the FIFO error.

It is likely that it is just after the place where interrupts are
re-enabled after being disabled for a while.
If this is splx() then you may need to dig further!  However I
suspect that is inlined in NetBSD.

If you have a serial console and ddb, just breakpoint on the code
that generates the message...

	David

-- 
David Laight: david@l8s.co.uk