On Sun, Oct 31, 1999 at 02:33:31PM -0700, Michael L. Hitch wrote:
> polled I/O - which I don't think is used by the device probes any more.

partially. That is, in my local sources I had to change it to use the polled
path if the data length is zero, to avoid the kernel panic....

> I think it's calling sbicgo() to do the select, and when sbicselectbus()
> returns with a select timeout, sbicgo() appears to assume it's trying to
> select a busy bus and ignores the select (assuming it will be scheduled
> again later).

Right. I think the polled path (called by my above change for INQUIRY, but
not for IDENTIFY) does the wrong thing, too, causing that the mid-layer to
call IDENTIFY anyway.

Can a sbic chip tell the difference between a busy bus and no device
responding? (I have a sbic manual somewhere, but... "somewhere")

Related question: Michael, and advice on whether sci.c should work, or needs
changes, too? (I could guess from the source, but you, at least, have once
written a sci backend...)

	-is