On Wed, 22 May 1996, Paul Goyette wrote:

> Are the via interrupts still in a state of flux?

That's not an entirely inaccurate statement.

>From what Allen and I have been seeing, I'd have to say that the ncrscsi
driver is depending on the older, broken behavior of the interrupt
handlers.  In effect, now that we're handling interrupts in the normal
(correct :-) fashion, the ncrscsi driver is missing the phase mismatch
condition at the end of a write.  It's conceivable, to me at least, that
something similar is happening with certain SCSI devices on reads, too.

There are two workarounds:  edit via.c to handle the VIA2 (or RBV)
interrupts the way that they are handled in the via1_intr() code, or use
the sbc driver in polled PDMA mode.  For the latter, you simply need to
configure it into your kernel and remove the ncrscsi driver; see the SPOT
config file.

This "bug" appears to be a hardware strangeness, and affects both the
ncrscsi driver and the sbc driver running in interrupt-driven mode.  Allen
and I are both working on it and sharing our results.

--scott

PS - if you use the sbc driver, your clock may drop ticks like crazy when
you access a slow device such as an MO drive or tape.  It works acceptably
well for me on my IBM Spitfire, though.