Subject: Re: serial ports, yadda yadda yadda
To: None <port-sun3@NetBSD.ORG>
From: "D. Hugh Redelmeier" <hugh@mimosa.com>
List: port-sun3
Date: 01/20/1997 13:16:16
| From: Ty Sarna <tsarna@endicor.com>
 ...
|   In fact, Sun
| serial ports are notoriously sucky, and if you've ever read the manuals
| for the Zilog chips that Sun uses, I don't think you'd call the 16550 a
| kludge -- it's really quite clean by comparison.
| 
| I'm running an ISDN net connection at 57.6 on a 25MHz 386 with
| reasonable success, while my 3/60 (20MHz vs 16 I think for the /180)
| couldn't keep up with 38.4 under SunOS.  I think NetBSD is better, but
| not by much. A Z85230 would likely help, but what it adds is the same
| thing a 16550 gives you: a larger FIFO. And it's not as large, and not
| as well designed, from what I gather.

I seem to remember hearing that there was a fair bit of care taken in
SunOS (at least the early versions) to get as good performance as
possible out of the Zilog chips.  I don't know if this "fell out" in
later versions.  My picture is that there is a tight "first level"
handler that effectively adds a software silo (aka FIFO).  This
impression is half-remembered from a 10 year old discussion with John
Gilmore at a party.

Is it easy to replace the UARTs in a 3/60 with a Z85230?  Does this
cause SunOS problems?  Does SunOS exploit it?  Does NetBSD exploit it?

I'd ask the same questions about my SPARCClassic, but I'm pretty sure
that the UART is integrated into a much bigger custom chip.

Hugh Redelmeier
hugh@mimosa.com  voice: +1 416 482-8253