Question on Target initiated negotiations...

Les Record 8-3622 lrecord at VNET.IBM.COM
Thu Jun 1 07:06:15 PDT 1995


> One proposal is to turn initiate on for wide negotiations based on
> the argument that most (all?) wide adapters must be more advanced
> and probably handle target initiated negotiations fine.  Is this
> a reasonable assumption?  Do people know of drives which do initiate
> negotiations by default?

There are indeed drives out there that initiate negotiations by default.
In general, they will wait to do it until it is absolutely necessary.  For
example, assume the drive has been reset for some reason, the initiator is
not aware the reset occurred (BDR from another host or a power-cycle of the
drive, for example), and the initiator issues a new command to it.  The
drive responds with a check condition (unit attention), and the initiator
issues a request sense without negotiating.  The drives I have in mind will
accept the command but then negotiate before entering the data phase.  The
bus sequence will look like this:

ARB
SEL W/ATN
MSGOUT - IDENTIFY
COMMAND - REQUEST SENSE
MSGIN - WIDTH NEGOTIATION (for wide drives)
MSGOUT - NEGOTIATION RESPONSE
MSGIN - SYNC NEGOTIATION
MSGOUT - NEGOTIATION RESPONSE
DATA IN
STATUS
MSGOUT - COMMAND COMPLETE

So far I have not seen any drives that will initiate negotiations before
the command phase, but there may be some out there.

My feeling is that all wide hosts must be prepared to accept target
initiated negotiations.  Narrow hosts should also be prepared for it but as
you indicated, some don't handle it very well.  Initiators should be
designed to renegotiate after any check condition status because they can't
know the details of the condition nor the state of the drive until they
retrieve the sense data.

Les Record  (512) 838-3622 Austin, Texas  USA
IBM Risc System/6000 Division -- I/O Verification
Internet: lrecord at vnet.ibm.com





More information about the T10 mailing list