Subject: Re: Another changer, another changer problem
To: NetBSD-current Discussion List <firstname.lastname@example.org>
From: Greg A. Woods <email@example.com>
Date: 10/08/1998 18:02:29
[ On Thu, October 8, 1998 at 13:38:49 (-0700), Curt Sampson wrote: ]
> Subject: Re: Another changer, another changer problem
> On Thu, 8 Oct 1998, Greg A. Woods wrote:
> > I realize what you're saying, but generally speaking most systems will
> > not have more than the maximum of four physical devices attached to any
> > given PCI bus....
> Huh? A machine right beside me has a five-slot motherboard, with
> two more devices on the motherboard itself. That's seven packages
> on this machine.
Well, I do know about some of the electrical specifications of PCI.
Unless someone's changed some hardware specs behind my back (and a quick
review at www.picmg.org suggests not), that either means you have at
least two separate PCI buses in your machine that are separated by a
bridge, or the mfgr has stretched the limit of the spec and you're not
likely to have a very reliable machine. So far as I know you can only
get to 6 or maybe 8 slots on the same bus if you go with CompactPCI.
However since you said "motherboard", you clearly are not refering to a
CompactPCI backplane. Ordinary PCI is limited to 4 slots per bus, which
is primarily an electrical limitation, BTW (and it seems 4 logical units
per slot, though I don't yet know much about this part of the spec.).
> Perhaps you could write up a detailed paper explaining in detail
> exactly how your proposed system would work. A large part of the
> problem I have with your scheme is I see a lot of hand-waving in
> the areas where I see serious problems.
Hmm.... You've not been very good at presenting the details of the
problems you see either.... You wave your hands in objection without
even really being the Devil's advocate. And you've been throwing in
objections based on systems where what I'm suggesting is clearly not
necessary or applicable. I hope you don't think I'm trying to say that
the status quo has to be completely obliterated. After all one of my
proposals is to implement boot-time hardwiring of devices, which would
imply little or no change in the current "look and feel" of the system.
However at this point in the discussion I'm still at the exploratory
phase, and for something like a revamp of the device configuration
datastructures into something other than full sized arrays I think we're
still some distance away from having things concrete enough to even
build an outline of the design, let alone writing up a detailed paper.
Of course some of the ideas floating around here are already at least
partly implemented in other systems, so if you want details I or someone
can give you URLs for their source code....
> > Perhaps a compile-time flag to select whether or not muliple controllers
> > per slot is supported or not would help here.
> Now we're back to compiling a new kernel again!
Yes, but only in easily predictable situations.
Greg A. Woods
+1 416 218-0098 VE3TCP <firstname.lastname@example.org> <robohack!woods>
Planix, Inc. <email@example.com>; Secrets of the Weird <firstname.lastname@example.org>