Re: Distinguishing identical devices

["John R Pierce" <pierce@hogranch.com>]

The usual convention is to sort them in bus/dev order and treat them as (for
example) scsi channel 0, 1, 2, ...  So the first instance of a SCSI class
controller would be scsi channel 0.  This is not unlike what Sun does on
their workstations, where the channels are searched in an arbitrary order
and the first one found is /dev/rdsk/c0xxxxxx, then /dev/rdsk/c1xxxxxx etc.

-jrp

-----Original Message-----
From: Mark Turner <turner@mc.com>
To: Mailing List Recipients <pci-sig-request@znyx.com>
Date: Friday, March 13, 1998 9:35 AM
Subject: Distinguishing identical devices


>We manufacture user-configurable multi-computer systems for
>high performance applications.  Our PCI version may have many
>copies of virtually identical compute node cards.  For several
>reasons we need to know which card is which.  These reasons
>might include the need to find an added-on I/O daughtercard, or
>the need to know where the board is connected to a secondary
>interconnect (which runs over the tops of the boards).
>
>Is there a commonly used way to distinguish between identical
>PCI cards in the same system?  The techniques that have come to
>mind are:
>
>  - Tell the driver software where each board lives in terms
>    of its bus#/device# on PCI.  This has some variability
>    depending on exactly how the system is configured, and
>    is not particularly user-friendly.
>
>  - Place in config-space somewhere (or in VPD?) a unique
>    identifier for each board, and have the driver software
>    read and make sense of this.  We'd still need to indicate
>    to the driver software something to relate what is read
>    to the system configuration as a whole.
>
>I imagine that there are systems that have, say, multiple SCSI
>controllers with disk chains off of them, or systems with
>multiple graphics systems (monitors) that need to be driven.
>How do these systems know which disk is which, or which screen
>is which?  Do people use user-settable switches that can be
>read by the driver software to distinguish devices?  Am I
>expected to go beyond the PCI interface to the device itself
>to ask it for some self-identifying information that can help
>to distinguish it?  Any suggestions or related anecdotes would
>be welcome.
>
>Thanks,
>
>Mark Turner
>Mercury Computer Systems
>turner@mc.com
>
>