Submission for comp-unix-microport

[Unix-to-Unix copy program]

Path: uclachem!cepu!trwrb!scgvaxd!wlbr!voder!pyramid!oliveb!ames!mailrus!umix!uunet!mnetor!spectrix!clewis
From: clewis at spectrix.UUCP (Chris R. Lewis)
Newsgroups: comp.unix.microport,comp.sys.ibm.pc
Subject: Re: Uport tape driver ev.o disables line printer
Keywords: device interrupts, memory windows
Message-ID: <471 at spectrix.UUCP>
Date: 1 Mar 88 02:51:05 GMT
References: <1206 at qetzal.UUCP> <446 at spectrix.UUCP> <243 at oracle.UUCP> <661 at anasaz.UUCP>
Reply-To: clewis at spectrix.UUCP (Chris R. Lewis)
Distribution: comp.unix.microport,comp.sys.ibm.pc
Organization: Spectrix Microsystems Inc., Toronto, Ontario, Canada
Lines: 70

In article <661 at anasaz.UUCP> les at anasaz.UUCP (L. Leslie Biffle) writes:
|In article <243 at oracle.UUCP> rbradbur at oracle.UUCP (Robert Bradbury) writes:
|>
|>
|>...The point was made that the IBM Micro Channel was designed with a much
|>different bus interface to get around this problem and allow exactly
|>what Chris Lewis suggests (namely to be able to chain device interrupts
|>at the same priority).
|>
|>I agree that there aren't very many interrupt lines on the PC, the
|>question is can you daisy chain them or do you need to get a PS/2?
|>
|
|Yes, the int request lines can be daisy-chained, though it takes
|thought on the part of the card designer.  Active-high vs. active-low
|is not the issue here as much as the physical execution of the
|hardware.  If the interrupt request line driver on an active-low bus
|is not an "open collector" device, the same conflict would exist where
|one device desires the signal to be high while another wants it to be
|low.
|
|Our intelligent data comm board uses a PNP transistor to drive the
|selected interrupt request line high when an interrupt is desired.
|When no board is driving the line high a simple resistor pulls the
|line low.  This resistor is enabled on the last-or-only board (in the
|spirit of the terminator on a disk drive).  In this way many of our 
|boards may share the same interrupt vector, and may coexist with other 
|boards that do the same.

Exactly.

Quite a few boards will actually work in such a scheme.  IBM hardware
treats the interrupt lines ALMOST like upside-down open collector gates -
except that there aren't any pull-ups (actually downs) on the mother
board.  The few real IBM boards I've examined, along with the one I designed,
don't use transistors.  They use tristate buffers (eg: 74125) who's 
input is hardwired to ground (or high in an inverting buffer) and you 
enable the buffer to fire the interrupt.  Further, you have to put
a pulldown somewhere....  Our solution for multiple cards with the
same interrupt was to put a resistor with the highest possible reliable
value on each board as a pull-down.  If you had only one board, the
pull-down was enough.  If you had three boards and three simultaneous
interrupts, the value was 1/3rd but 74125's have far more than enough 
drive.  We did it this way to eliminate another (error-prone) jumper.

Potential problems:
	1) you have to be interrupting on *level* not edge.  (I know
	   386/ix uses level - as do most other UNIXes if they have a choice, 
	   don't know about messydos)
	2) You may be sharing the interrupt with some board that doesn't
	   drive the line properly.

One of the nastier things that arise out of this STUPID design of IBM's
is that since interrupt lines aren't pulled inactive, you can sometimes
get continous interrupts on a line that isn't connected.

What makes this worse is that 386/ix's config (uPort too I imagine) does
not notice interrupt clashes, nor do they do the "right" thing and
create a dummy ISR that calls all clashing *real* ISRs.  Multibus I
has very few interrupts (if you're using the vast majority of boards
that can only autovector but not bus vector) and most systems on Multibus
will create a dummy ISR (eg: Tower and Spectrix machines).

I'm glad that no IBM'ers were around when I found out this stuff the
hard way. GRRRRR!!!!!  STUPID IDIOTS!  Ruined a perfect first-run of the 
PC board....
-- 
Chris Lewis, Spectrix Microsystems Inc,
UUCP: {uunet!mnetor, utcsri!utzoo, lsuc, yunexus}!spectrix!clewis
Phone: (416)-474-1955