QErr bit and Multi-Initi

Charles Binford chas at smtplink.wichitaks.ncr.com
Tue Aug 16 14:53:13 PDT 1994


Part of my (Charles Binford) original comments: >>
Jim McGrath's reply: >

>>    What I don't understand is what prevents two initiators from continuously
>>    aborting each other's I/Os?  Consider the following:

>What prevents this is quality devices.  CHECK CONDITIONS are not usual
>occurances.
>At least with disk drives, you are very unlikely to ever see the situationn
>of
>CHECK CONDITIONS all the time screwing up two devices.  If you see that,
>then there is something fundamentally wrong at the target side (e.g. the
>device is NOT READY), and nothing you do at the SCSI interface will help
>one way or the other.
>
>Jim

I agree Jim, CHECK CONDITIONS are not normal occurances.  My point is that with 
multi-intiators, ONE check condition to one of the initiators has the potential 
to snowball into nearly continuous check conditions because of the unit 
attention returned to the "other" initiator.  It seems to me that SCSI defines 
circular logic here.  The mechanism used to inform the other initiator that its 
I/Os were aborted because of a CHECK CONDITION, generates another CHECK 
CONDITION.

I seems to me that the QErr bit would work much better if it only aborted I/Os 
|from the initiator which saw the CHECK CONDITION.
Charles Binford






More information about the T10 mailing list