CLARIFICATIONS

Tom Wicklund wicklund at intellistor.com
Wed Aug 17 08:33:48 PDT 1994


GFRAZIER at ausvm6.vnet.ibm.com writes:
 > Just a couple more clarifications to recent e-mail:
 > 
 > 2) Tom Wicklund wrote:
 > 
 >  In the same way, I don't see why a SCSI-3 device might not accept
 >  selection without ATN as a signal to revert to SCSI-1 operation.  It
 >  might fail a strict SCSI-3 compliance tester, but won't affect normal
 >  operation on a "real" system.
 > 
 > Yes, there will be no problem under "normal" circumstances, however,
 > as a previous note says:
 > 
 >  "... if the targets accept commands without an IDENTIFY
 >  message, they expose themselves to data integrity problems because they
 >  will route all commands from SCSI-3 initiators to LUN 0 whenever the -ATN
 >  wire is broken or it's driver is defective."

This same problem exists in SCSI-2.  SCSI-2 recommends that the LUN
field be set to 0 and the IDENTIFY message used to specify the logical
unit.  If the ATN line is broken in a SCSI-2 system the same data
integrity problems exist.

The real question is what a SCSI-3 target should do when selected
without ATN.  SCSI-2 says the IDENTIFY message must be sent, but
leaves the action of select without ATN as reverting to SCSI-1
compatibility.  SCSI-3 should specify what to do if ATN is not raised
as required.  Requiring that the target go to BUS FREE (or perhaps
allowing it to revert to SCSI-1 operation) seems reasonable.


The real responsibility for error detection is in the initiator.  The
initiator can determine that the target has not entered MESSAGE OUT
phase after selection and reset the device (since ATN wasn't
recognized, it can't use an ABORT or BUS DEVICE RESET message).

My guess is that an initiator which blindly sends things out with a
broken ATN line would send the IDENTIFY message anyway, which the
target would interpret as the first command byte, and an error would
be detected.




More information about the T10 mailing list