Faster Serial Ports... How can I?

[Rick Richardson]

In article <4540 at b-tech.UUCP> zeeff at b-tech.UUCP (Jon Zeeff) writes:
>In article <691 at omen.UUCP> caf at omen.UUCP (Chuck Forsberg WA7KGX) writes:
>>
>>Replacing the 8250 or 16450 UART ships with the new National NS16550AN
>>allows Pro-YAM to activate a 16 deep FIFO register.  This increases the
>>allowable interrupt latency by almost an order of magnitude.  It is
>
>Perhaps Microport could consider supporting this chip in their asy driver.
>It's probably a simple change.

(Sorry, Jack, the cat is out of the bag now!  Might as well spill my
	guts too!)

Two of us have been running these parts (the 16550A) since early February
in our old AT's running the 'other' UNIX: Venix System V/286.  We both
have upgraded to 386's, but my 286 is still doing yoeman service
handling the Trailblazer at 19.2 with the 16550A -- I could see little
reason to upset the stable, solid 286.

On the surface the change is a one-liner.  Just enable the FIFOs.
In fact, before hacking the source I tried this approach by
writing a little application that root could execute to turn on
the FIFOs.  Could have been a net-hero if this had worked!!!!!

However, this did not simply work in Venix, because the interrupt
service routine tried to process interrupts until the IIR register
was equal to "1".  With FIFO's enabled, this register will never
contain that value.  The original test looked like this:

	while ( (x = inb(IIR)) != 1 )

when it should have (properly) been:

	while ( ((x = inb(IIR))&1) != 1 )

I don't know if other UNIXes have this same problem.  If they do,
you'll need source, and/or a desire to squeeze the AND instruction
into the existing object code.

To enable the FIFO's is really straightforward.  You have to decide
whether you want to A) minimize interrupt overhead at the expense
of allowable interrupt latency, or B) maximize allowable interrupt
latency at the expense of interrupt overhead.

For the 286, the answer is "B".  You want 16 deep FIFOs that
trigger as soon as any characters are available.  That gives you
the most breathing room for sloppy kernel coding in other areas.
The 16550A FIFO control register is a write only register at
the same address as the 8250 IIR register.  Just write a "1"
into that register and whammo! the FIFOs are on.
-- 
		Rick Richardson, President, PC Research, Inc.

(201) 542-3734 (voice, nights)   OR     (201) 389-8963 (voice, days)
uunet!pcrat!rick (UUCP)			rick%pcrat.uucp at uunet.uu.net (INTERNET)