SCSI Tagged Queueing And Autosense

Alan Spalding (714) 932-5916 SPALDING_AL at a1.WDC.com
Fri Dec 22 18:49:00 PST 1995


* From the SCSI Reflector, posted by:
* <"Alan Spalding (714) 932-5916" <SPALDING_AL at A1.WDC.COM>>
--EiqaQBNL8KlZ2SbwmFQ24B==
Content-type: TEXT


--EiqaQBNL8KlZ2SbwmFQ24B==
Date: Fri, 22 Dec 1995 01:38 PST
Subject: SCSI Tagged Queueing Vs Autosense Question
Sender: "wdcscsi-owner at dt.wdc.com" <WDCSCSI-OWNER at DT.WDC.COM>
To: "WDCSCSI at dt.wdc.com" <WDCSCSI at DT.WDC.COM>
Content-type: TEXT
Posting-date: Fri, 22 Dec 1995 07:00 PST
Importance: normal
A1-type: MAIL




--vqXr+KViNW3RYAltqcHRJD==
Date: Fri, 22 Dec 1995 00:43 PST
Subject: Tagged Queueing Vs AutoSense question
Sender: "Alan Spalding (714) 932-5916" <SPALDING_AL at A1.WDC.COM>
To: Alan Spalding <SPALDING_AL at A1.WDC.COM>
Content-type: TEXT
Posting-date: Fri, 22 Dec 1995 01:36 PST
Importance: normal
A1-type: MAIL

Hi,
I'm not particularily familiar with Tagged queueing and would like make sure I 
have understood the SCSI-2 spec correctly.

A specific question that I have is how a SCSI HBA should implement autosense. My 
first thought was that when a check condition is reported an untagged request 
sense should be issued. But my understanding of the spec is that when this R.S.   
has completed any queued commands will be restarted (unless QErr bit is set to 
0). This would seem to be incorrect for certain error conditions such as 
replaced media - maybe the device aborts I/Os when the media is replaced). Is 
the only safe algorithm to use:

on receipt of the check condition - issue either an abort tag msg for each I/O 
in the queue or a clear queue message prior to issuing the untagged R.S.? 

Am I correct in my assumption that the first untagged command  restarts the 
suspended queue and that any tagged commands issued prior to the first untagged 
command  will simply be queued in the device queue?

Regards

alan spalding (internet spalding_al at a1.wdc.com)  

--vqXr+KViNW3RYAltqcHRJD==


--EiqaQBNL8KlZ2SbwmFQ24B==




More information about the T10 mailing list