From owner-freebsd-hackers  Sun Mar  2 23:47:23 1997
Return-Path: <owner-hackers>
Received: (from root@localhost)
          by freefall.freebsd.org (8.8.5/8.8.5) id XAA08503
          for hackers-outgoing; Sun, 2 Mar 1997 23:47:23 -0800 (PST)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.8.5/8.8.5) with ESMTP id XAA08498
          for <freebsd-hackers@FreeBSD.ORG>; Sun, 2 Mar 1997 23:47:20 -0800 (PST)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.8.3/8.6.9) id SAA18806; Mon, 3 Mar 1997 18:43:10 +1100
Date: Mon, 3 Mar 1997 18:43:10 +1100
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199703030743.SAA18806@godzilla.zeta.org.au>
To: ejs@bfd.com, tom@sdf.com
Subject: Re: sio projects
Cc: bde@zeta.org.au, freebsd-hackers@FreeBSD.ORG
Sender: owner-hackers@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

>> I'd be surprized if the fifo's are actually disabled, as it's hard to get
>> unix to do better than 19.2kbaud without them.  I suspect that the
>
>  No.  FreeBSD handles 57600 quite nicely on a 16450, which only has a
>single byte buffer.  At 115200, you may loose a few characters, but its
>quite usable.

It shouldn't lose any even at 115200, at least with matched UARTs and
working flow control, except maybe on a 386SX/16 or with a bus-hogging
(broken) bus-mastering DMA controller.  My standard test is to run 2
channels at 115200 bidirectional saturated through 16450's on a 486/33
ISA system with a slow IDE drive.  This generates almost 4 * 115200/10
interrupts/second and consumes 75% of the CPU.  However, there is a
factor of about 4 to spare (at a cost of reduced throughput) provided
flow control is working.  Setups with mixed (active) 16550's and 16450's
don't work so well.  It is possible for the 16500's to starve the 16450's.
However, it probably takes more than one 16550 to do this on a 486/33.

Bruce