Subject: Re: 486 with 56K modem?
To: Bruce Anderson <BruceA@SpaceStar.Net>
From: Aidan Cully <>
List: port-i386
Date: 11/14/1998 14:30:01
On Fri, Nov 13, 1998 at 06:26:52PM -0600, Bruce Anderson wrote:
> Note -Remove that "call isp"
> and use 
> # pppd call isp
> instead to bring the link up.

I'll do that when I have to worry about having more than one ISP at
once..  It's a fairly trivial change.

> BA-
> DX33 with 1.3.2 and Zoom 56K

I changed the speed from 115200 to 57600, added an mtu 512 line, and
changed my mru from 542 to 512..  The NetBSD box hasn't crashed, yet,
so I guess it's something better, but we still get all kinds of errors
on reads.  I see a couple of messages like this from the kernel:
com0: 8 silo overflows, 0 ibuf floods
com0: 7 silo overflows, 0 ibuf floods
It looks like the 486 just can't keep up with the interrupts we get
from the modem..  This could very easily be the VLB, since we've got
three VLB cards in the thing (including the IO controller with the
serial ports), and I understand a 66MHz machine is only supposed to
be able to handle two.

> My /etc/ppp/peers/isp looks like this:

It looks like there are very few functional differences between
your options file and mine..  Anyways, I seem to have found a way to
make pppd connect without crashing the box.

The reason I sent this to port-i386 is that Linux and NetBSD look to
have different philosophies for reading from com ports, and Linux
works while NetBSD doesn't for us.  It looks to me like Linux doesn't
rely on hardware generated interrupts from the 16550 at all, it just
polls the device, while NetBSD never polls, always uses interrupts,
and I thought that alone _could_ account for the fact that the modem
works on Linux, but fails on NetBSD.  Now I'm fairly sure the problem
is a VLB that can't handle sending 1800 interrupts a second to the
kernel, and I'm going to try to snarf an ISA card from somewhere to
handle serial/disk IO.