Date   : Wed, 27 Jul 2005 20:12:35 +0200 (BST)
From   : Johan Heuseveldt <johan@...>
Subject: Re: Econet-Ethernet bridge

Hi Jules,

On Wed 27 Jul, Jules Richardson wrote:
> On Wed, 2005-07-27 at 15:50 +0200, Johan Heuseveldt wrote:

> > Iyonix computer doesn't have a parallel port anymore. New printers also
> > seem to come without one (only USB and/or network). 
> 
> From a hacker point of view that's a shame :-(
> 
> > So, Econet for Beebs is at max 200 kHz, which is bit speed. Byte speed
> > will then, lets say, 20 to 25 kHz. (20/10, 20/9, 20/8)
> > Is that (really) a problem for parallel/Centronics?
> 
> I *think* you can get about 100KB/s through a parallel port, so it
> sounds like that *could* work comfortably with buffering at the byte
> level.
> 
> Just depends whether it's possible to work at the byte level across that
> interface boundary - which depends on the nature of the Econet protocol
> itself (if it needs to work at the bit level at any time then control
> via the parallel port - operating at byte level - might be difficult)

Not sure if I understand you correctly, but the standard software
(NFS ROM) always accesses the 68B54 at byte level for control, status
and data, as the there is no other way of accessing the device.

There are two other hardware addresses - which don't access the 68B54
- and allow/disallow the NMI from the 68B54 to the NMI line. They are
called 'notINTOFF/notSTATID' and 'notINTON'.

The NFS software also uses this to do some processing while servicing
the NMI interrupt, and polling the device when it suits. And it is used
to switch off NMIs completely, if someone else claims (and gets) the
use of NMIs, like DFS.



Hm, that makes me wonder. Perhaps the interrupt issue is more important:
Is it possible to deliver such an NMI interrupt from the 68B54 fast
enough to the PC system, and can the system respond quickly enough on
that, without over/under-run problems at the 68B54?
The first four/six bytes are not just data, but needs processing as well.
(stations/nets/port/flag bytes)

In other words, could there be the same problem as with the EcoLink
cards? They had an processor on board to overcome the problem of
interrupt latency. The 68B54 only has 3 bytes buffers for its data to
receive/transmit.
(the 6502 is pretty fast in response to an interrupt, and is quite
unique as such)


The reception of a data frame starts with interrupts per byte, and
processing is needed to determine if the packet is for this machine.
For the actual data, it is switched to two bytes per interrupt.
(the Tube ULA can also be switched to such a two byte mechanism)

No doubt machine (PC) and parallel connection are on average quite
fast enough, but how about interrupt latency? 

Can a NMI-like interrupt line be used, originating from the parallel port?

The more I think about it, the more problems I see coming up. :-(


greetings,
Johan

-- 
Johan Heuseveldt <johan@...              >
  aka  waarland

  The best place is a Riscy place
 
"... a sheep in sheep's clothing ..." - Winston Churchill