Subject: Re: Another changer, another changer problem
To: NetBSD-current Discussion List <current-users@netbsd.org>
From: Greg A. Woods <woods@most.weird.com>
List: current-users
Date: 10/08/1998 21:45:13
[ On Thu, October 8, 1998 at 16:54:53 (-0600), Justin T. Gibbs wrote: ]
> Subject: Re: Another changer, another changer problem 
>
> >Well, I do know about some of the electrical specifications of PCI.
> 
> The spec, as I know it, is 10 loads per bus.  A slot is considered
> two loads.  This is the reason why you usually only see 4 PCI slots
> in a system.

I don't know which specification you might be thinking of, but the PCIMG
PCI/ISA specification only allows for 4 PCI slots.  Period.  There are
only four clock signals, only four grant signals, and only four request
signals; one of each specified for each slot.
 
To quote from the URL I posted previously:

	The specification defines backplane signals for up to four PCI
	bus slots [....]

Perhaps there are other specs that allow more slots, but my reading in
various trade journals has revealed that some/most/many experts feel
that putting any more than four devices on a PCI bus will risk serious
reliability problems too.  My understanding of the electrical behaviour
of a high-speed bus agrees.  (Wait 'til you see how a prorietary design
for a 350MHz, 64-bit VME bus works!  They're supposedly in production!)

The specifications for the 2115[0234] PCI-to-PCI bridge chips also show
that 4 devices per PCI slot is the maximum.  Of course there's nothing
preventing buses from being chained together with such bridges, so each
card might have as many as 9 devices (eg. host adapters, or NICs) on
board (that's probably where you get your "10 loads per bus" number).

CompactPCI (also as defined by PICMG) allows for a maximum of 8 slots on
the backplane (without having a PCI-PCI bridge), and of course one of
those slots it always the "system" slot for the SBC.  (This is one of
the reasons that operating systems which support PCI based motherboards
may not instantly also support CompactPCI systems containing otherwise
nearly equivalent hardware.

-- 
							Greg A. Woods

+1 416 218-0098      VE3TCP      <gwoods@acm.org>      <robohack!woods>
Planix, Inc. <woods@planix.com>; Secrets of the Weird <woods@weird.com>