Subject: Re: NetBSD/i386 and single board computers
To: None <port-i386@NetBSD.ORG, port-i386@NetBSD.ORG,>
From: Matthias Drochner <firstname.lastname@example.org>
Date: 12/04/1997 18:24:10
Excerpts from netbsd: 3-Dec-97 Re: NetBSD/i386 and single .. Greg A.
> We had 1.2 or thereabouts running on both P90 and 486 single-board
> computers (one of each on a split backplane actually).
I have about 20 single board computers running now, PICMG
PCI/ISA from 4 different manufacturers. They are
completely PC compatible, NetBSD is running without problems
on all of them.
My experiences with split backplanes are worse - epecially
with these "power sequenced" backplanes which turn on the
slots individually. I experienced power shortage when the slots
drew too much current (there is 1 power supply for both systems),
and I even managed to smoke some of the switching FETs.
Besides this, the switching logic was junk in my case, it happened
that the CPU got power seconds before the slots and hung
immediately because it tried to configure powered-down
PCI slots. Switching off and on after less than 10 seconds
confused them completely.
A general design problem is that the CPU cannot be reset
from the backplane.
My experience is that the systems become unstable if there
are more than 3 PCI slots, especially for demanding and
DMA-intensive use. Look at the trace length between
the PCI master slot and the slaves...
When I tried a backplane with 16 PCI slots (4 PCI-PCI bridges)
it was a complete disaster: No 1 slot was properly configured
by any of the CPU boards. Obviously the backplane uses
an IDSEL routing which the CPUs don't expect.
BIOSes are generally a weakness: They are generic PC
BIOSes, not well suited for embedded use. (One exception:
A Teknor CPU could be partly managed over a serial line.)
It's a bad joke to have an AMI WinBIOS on an "industrial"
There are more things to complain about - missing CPU
fan power sockets (besides the fact that fans cannot be
trusted in a rugged environment anyway), and CPU cards
which don't fit into 64-bit slots, and no standardized way
to boot from flash ROM, and ... ... ...
> I don't see why there'd be any more
> problems than with your average PC hardware! ;-)
PCs are junk regardless of the shape they have.
> Looks like you can build a decent
> computer out of CompactPCI equipment and get good performance to boot,
> and not suffer without driver support
I've got 2 CompactPCI systems too. I suspect the PCI
bus to be electrically unstable, at least I'm experiencing
random hangs if a device does DMA. The logic analyzer shows
that the device properly asserts the bus REQ, but never gets
The Intel PCI chipsets can only drive 4 PCI masters,
one would assume that a PCI-PCI-bridge is necessary
to connect the outside bus. However, there is no - they
probably did bad tricks with the clock and arbitration lines...
I'd probably trust the FORCE boards a bit more, but they
refused to sell me one without enclosure, disk and NT
I never had this amount of trouble when I worked
with VME boards. But they never had an 80x68, an
ISA bus nor a PC BIOS...