Johnathan Vail vail at
Tue Aug 16 13:28:58 PDT 1994

Date: Tue, 16 Aug 94 11:01:31 CDT

   With reference to my proposal to require SCSI-3 targets to free the bus if
   -ATN is not activated after selection, Gerry Houlder wrote:

     "This change means that SCSI-3 targets cannot function in SCSI-1 or
     SCSI-2 systems that use selection without ATN. This will absolutely
     require initiators to put the drive in "SCSI-2 mode" or "SCSI-3 mode"
     before installing it in the system. ..."

   A tradeoff needs to be made between compatibility with very old systems and
   data integrity. I may be wrong, but I would guess that there are not
   many very old initiators which do not activate the -ATN line, and
   these initiators are probably not in systems which contain SCSI-3

I write firmware for SCSI targets and initiators.  Since most of my
applications are in closed special purpose systems I simplify things
and don't use SELECT with ATN for most commands.  With SCSI-3 I can
update my firmware easily but I hate new protocols that break older

   Another recent comment asked why the -ATN line is singled out for
   protection. It is singled out because it is unprotected and it can cause
   loss of customer data. The data lines, which obviously can cause loss of
   customer data, have been protected by parity since SCSI-1. With a higher
   percentage of multiple LUN devices becoming available, it's time to protect
   against -ATN failures also since these failures may be equally

If an INITIATOR SELECTs with ATN the next phase should be MESSAGE OUT.
Anything else and the transaction is broken by a phase mismatch.
Wouldn't this detect a broken ATN line?

A mandatory MESSAGE OUT following a SELECTION would only serve to
ignore the ATN line and hide bad hardware until later.

Or am I missing something still?


PrePRESS      Johnathan Vail     vail at     (508) 663-7435
---------+    900 Middlesex Turnpike   Billerica Massachussetts   01821
solutions     MEMBER: League for Programming Freedom (league at

More information about the T10 mailing list