Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!eecae!netnews.upenn.edu!rutgers!bellcore!faline!thumper!ulysses!andante!alice!debra
From: debra@alice.UUCP (Paul De Bra)
Newsgroups: comp.unix.xenix
Subject: Re: Need info on 16-bit DMA for AT
Keywords: 8237-5, AT Bus
Message-ID: <9082@alice.UUCP>
Date: 21 Mar 89 17:15:49 GMT
References: <1901@randvax.UUCP> <37068@bbn.COM> <1751@csuna.csun.edu> <13366@steinmetz.ge.com> <1774@csuna.csun.edu>
Reply-To: debra@alice.UUCP ()
Distribution: usa
Organization: AT&T, Bell Labs
Lines: 40

In article <1774@csuna.csun.edu> abcscnge@csuna.csun.edu (Scott Neugroschl) writes:
>In article <13366@steinmetz.ge.com> davidsen@crdos1.UUCP (bill davidsen) writes:
>]In article <1751@csuna.csun.edu> abcscnge@csuna.csun.edu (Scott Neugroschl) writes:
>...
>2.2.1, most notably during getty.  Also during ASCII uploads (9600/7/1/odd)...
>Loses console and serial interrupts during disk access... Maybe I'm assuming
>the wrong reason, but I've had friends who wrote device drivers tell me that
>the disk is run from the CPU rather than DMA, and with interrupts disabled...
>
>Anyone familiar with Xenix Sys V 2.X internals know anything?
>
>P.S.  This is on a True Blue AT339 (8MHz, 1MB (counting the genuine Intel
>Above Board), ST-4038 HD.
>

I have done quite a bit of experimenting on an AT (8 Mhz and 10Mhz but that
makes little difference) with 2.2.1.
I have never had my system loose input from a serial line, even with 2 9600
baud incoming uucp connections. (i.e. no "alarm" messages) This in contrast
to my new 25Mhz 386 box which looses interrupts with 1 incoming 19200 baud
line if you attempt to do anything else.

The serial interrupts normally have the highest priority. The only way I
found to cause the serial lines to miss interrupts is to lower the priority
and then cause some heavy activity on other devices. I wrote my own drivers
(well, with a friend) for sound, for slow parallel printers, for ramdisks
and none of this had any influence on the serial lines either or vice versa. 
An incoming uucp did not slow down the prelude by Bach being played on the
speaker. The only thing that might cause problems and occasionally caused
a glitch in Bach is switching between virtual consoles. That seems to take
a "long" time while blocking interrupts.

In any case the serial interrupts are handled much more efficiently in
Xenix than in any other Unix implementation for the 286/386.

Paul.
-- 
------------------------------------------------------
|debra@research.att.com   | uunet!research!debra     |
------------------------------------------------------