X3T9.2/93-016R0

To:       Membership of X3T9.2

From:     Ralph O. Weber
          Digital Equipment Corporation

Date:     January 21, 1993

Subject:  Why SCSI-3 needs ABORT (all initiator tasks)


At the January Working Group meeting, consensus was reached regarding the
absence of any need for an ABORT (all tasks from this initiator) function in
SCSI-3.  (Note: this function is equivalent to that performed by the ABORT
message in SCSI-2.)  The agreement was based on the idea that a series of
properly formed ABORT (one task) operations can produce the same effect as
an ABORT (all initiator tasks) function.  (ABORT (one task) is equivalent to
ABORT TAG.)

I believe that an appropriate use DOES exist for the ABORT (all initiator
tasks) function.  The use occurs during initiator crash/reboot sequences. 
In such cases, the initiator lacks sufficient information to construct well
formed ABORT (one task) operations.  In fact, the initiator does not know how
many ABORT (one task) operations are needed.

In the absence of an ABORT (all initiator tasks) function, the initiator
engaged in a crash/reboot sequence must use one of the "bigger hammer" func-
tions, such as CLEAR TASK SET (equivalent to CLEAR QUEUE) or TARGET DEVICE
RESET (equivalent to BUST DEVICE RESET).  These functions have consequences
very much beyond the actual needs of the initiator's crash/reboot sequence.
Using these functions in multi-initiator environments, will result in a
crash/reboot sequence on one initiator producing unnecessary lost-task
recovery activities on all other initiators connected to the target.

With this in mind, I propose the following list of functions having task
killing capabilities and appropriate uses of each function:

  ABORT (one task)      User requested termination of an operation.
                        Initiator controlled recovery from bounded software
                        failures.

  ABORT (all initiator  Recovery during an initiator crash/reboot sequence.
  tasks)

  CLEAR TASK SET        Recovery from some device failures (implies
                        initiator knowledge of how the target can be managed
                        to achieve a useful recovery).

  TARGET DEVICE RESET   Recovery from other device failures, where the
                        initiator wants the target returned to a known
                        baseline state.