> I was seeing "silo overflows" when doing "cvs update" over a dial-up
> line on my 486, from 1.6F to 1.6J-2002-10-28 (and I never saw that on
> 1.5.2), but they're gone after updating the kernel to yesterday's
> sources.

You could try to debug the problem!

There are two likely senarios:

1 If interrupts have been disabled too long, then the interrupt
  that generated the 'silo overflow' message is (likely) to be gennerated
  from an interrupt that happens just after the splx() call that
  enables the serial interrupt.

2 The serial ISR is spending too long doing something other than
  removing characters from the RX fifo.

Getting kdb to breakpoint on when the message is output might
give 'interesting' stack traces.

	David

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