PnP SCSI Industry Meeting, Jan. 10, South Lake Tahoe

Steve Timm (SYS) stevetim at microsoft.com
Tue Feb 7 14:54:36 PST 1995


This industry meeting was called to discuss and identify any changes
required to update the Plug and Play SCSI Rev. 1.0  spec. dated
March 30, 1994.  The review was prompted by feedback from implementers
following a recent PnP SCSI Plugfest and updates to Parallel SCSI-3 SPI.
In addition, new proposals for sustaining or extending PnP SCSI's definition
and applications were discussed.

ATTENDEES: Hyundai MPD (formerly NCR MPD); Adaptec; Conner; Digital Equipment;
Quantum; IIX; Zadian Technologies; Seagate; Sun Microsystems; Unitrode; 
IBM; HP; Future Domain;
FSI; Lawrence Livermore; Qlogic; Hitachi; Burr Brown; Dallas 
Semiconductor; Apple; Tandem;
Western Digital; Fujitsu

SUMMARY
Much of the discussion centered around design recommendations and
clarification of the SCSI-3 SPI Annex B "SCAM" protocol spec. Results include:
 - Discussion and recommendations for PnP SCSI implementation including
    considerations for support of a 25 pin connector, 16 bit wide SCSI, 
and PCMCIA SCSI.
 - Agreement on SCAM protocol design recommendations for submission to
   X3T10 for approval and subsequent update to SPI Annex B.  Specific items
   for submission include input from Conner and Quantum summarized below.
   Most of the recommendations provide excellent clarification for PnP SCSI
   design.  Implementers are advised to consider these recommendations
   for their designs until final requirements are approved for 
inclusion in SPI.


RECOMMENDATIONS for UPDATE of  PnP SCSI 1.0 :
1.  PnP SCSI requires use of a standard external 50 pin high density 
connector.  PCMCIA cards
     however are limited to a 32 pin cable connector adapting to a 50 
pin high density, possibly
     violating PnP SCSI.  Recommendation is to require bundling the 
bridge cable for PnP compliance.
2.  Proposal to define an optional low cost PnP SCSI profile based on a 
25 pin connector was discussed
     and deferred.  Most felt that a second connector option weakens 
PnP SCSI, but growth of PnP into
     the high end and low end markets may warrant further review.  
Other related PnP connector issues
     discussed included: mixing narrow and wide SCSI and insufficient 
grounding for 25 pin fast SCSI timing.
3.  Some CD-ROM drives do not supply termination power and thus do not 
work when attached to a
     PCMCIA SCSI card (reference section 6.2 of PnP SCSI version 1.0).  
Recommendation: Low power
     battery operated PnP SCSI peripherals must supply termination power.
4.  ID assignment to mixed bus width devices (ie; 8 bit external, 16 
bit internal) needs clarification for
      consistent design.
5.  Industry/X3T10 and Microsoft ownership of PnP SCSI spec. was raised 
as an open item.
     Microsoft's recommenation is Industry ownership as a technical 
report under X3T10.   X3T10 Action is
     needed to publish this report.  (Note: Microsoft currently 
references the PnP SCSI profile as the
     basis for logo certification of SCSI products designed for Windows 95.)

SCSI-3 SPI ANNEX B Revision 15 SCAM Protocol Recommendations:

CONNER   SCAM DESIGN RECOMMENDATIONS
Note: Items 2,
1.  Various editorial comments
      - Section B4.1.1 item 4 should read: "Read and latch data from 
the DB(4-0) signal.  All devices assert
         the DB(6) signal.
      - Section B4.1 rewrite is needed to clarify use of vague terms 
such as "device" when this may
         be either a target or initator or a SCAM or non SCAM device.  
Also need some clarification
         on when to assert the MSG signal.
      -  Use of SCSI terminology such as SCAM Initiator or SCAM Target 
instead of master or slave.
2.  Section B5.1.2 should mandate a 32 byte ID string to avoid 
incompatible implementations.
3.  Section B5.1.3:  SCAM initiators and targets should support the 
"Configuration Process Complete"
      function code to close out ID selection until the next reset.  
This function code is strongly encouraged
      to avoid potential ID conflicts such as the rare case of more 
than 7 attached peripherals.
4.  B.6.1.1  Polling of the bus by a SCAM initiator by attempting a 
selection should be followed by
      a specific command such as Inquiry or Test Unit Ready to close 
out the selection gracefully.
      Recommendation is to choose one of these commands and make it a 
requirement.
5.  B5.1.1: Wired-OR glitch filtering must be implemented properly to 
ensure interoperability.
6.  Section B5.3.1: Some SCAM targets do not support some action codes 
such as "locate on" and
      may stop participation in SCAM selection after receiving this 
action code even though they have not
      yet received an "assign ID" action code.  Drives must not 
misbehave in response to unsupported
      funciton/action codes.  Target support for all function/action 
codes is recommended.
7.  Timeouts and error reporting for common bus hang conditions is 
lacking in the current specification
      and is recommended for enhanced PnP SCSI design.  Additional 
standardization is required here.
8.   B6.1.1.1 states SCAM tolerant reset to selection delay is 250ns.  
Failure of the initiator to honor this
       delay can result in failure to acknowledge a SCAM tolerant 
legacy device resulting in ID conflicts.
       This delay is mandatory to ensure interoperability.

QUANTUM SCAM DESIGN RECOMMENDATIONS
 1.  SCAM initiators must execute SCAM selection in response to a bus 
reset since some SCAM
      peripherals revert to their ID assignable state following a 
reset.  Failure to reassign IDs following
      a reset condition can result in an ID conflict or the inability 
to address devices on the SCSI bus.

1.1  Also, if a SCSI bus reset does occur during SCAM ID assignment, 
set IDs are  cleared and
      on completion of SCAM (assuming it is not restarted in response 
to reset) some devices
      may revert to their original ID.  Therefore, it is recommended 
that bus resets not be prohibited by the
      firmware or SCAM device drivers during SCAM selection.  For new 
designs, it is recommended that
      SCAM devices not default to their hard ID on SCSI reset.



Note: A followup Plug and Play SCSI plugfest is being planned for 
April/May timeframe
            Please send mail to stevetim at microsoft.com if you have 
questions about this event.
 - 




More information about the T10 mailing list