X3T9.2/91-19 Rev. 0
 
Date:   Feb. 15, 1991
 
To:     X3T9.2 Committee (SCSI)
 
From:   Paul Hanmann
        Emulex Corporation
        3545 Harbor Blvd.
        Costa Mesa, Ca. 92626
        Work phone: (714)668-5328
        Internet:   p_hanmann@emulex.com
 
Subject:  Using MODE SELECT to reset a drive to defaults.
 
Before SCSI-2 became an accepted reference for designing SCSI products, people 
used the CCS standard and SCSI-1 as the definitive answer on how to design 
SCSI devices. One of the problems was the way in which a peripheral device, in 
my case a disk drive, could be reset to defaults. As a OEM host bus adaptor 
manufacturer, most of the time we have little or no control over the 
configuration of drives which are hooked to our HBA's originate.
 
The accepted method of accomplishing this was through the following sequence:
 
1.      Request Default MODE SENSE pages from the drive.
2.      Request Changeable MODE SENSE pages from the drive.
3.      Mask to save only the changeable values.
4.      Make changes to changeable fields for Initiator required values.
5.      Send back pages thru a MODE SELECT.
 
This accomplished the function of  setting the drive "back to a known state".  
This is supported by the CCS spec and can be found on page 20, second 
paragraph:
 
"The Target shall create a CHECK CONDITION status when receiving a non- zero 
value in a MODE SELECT field or bit that the target declared non- changeable 
in MODE SENSE data."
 
SCSI-2 appears to have not addressed the problem at all but if anything has 
reversed the direction taken in CCS by the following found on page 7-31 of 10c 
in the paragraph right after the Implementors Note:
 
"The target shall terminate the MODE SELECT command with CHECK CONDITION 
status, set the sense key to ILLEGAL REQUEST and set the additional sense code 
to INVALID FIELD IN PARAMETER LIST for the following conditions:
1) If the Initiator attempts to change any field that is not changeable as 
reported by the target. In this case, no parameters shall be changed by this 
command."
 
This appears to be the exact opposite of the CCS standard set and used by many 
for years. I have no problem with this method and would be glad to just 
implement software if this was the only problem.
 
In testing we have found that a number of manufacturers currently allow non-
changeable values to be sent back as zero, echoed default values, and even 
incorrect values(ignored?). This causes many problems in coding software to 
handle all circumstances and is more complicated the deeper you go.  Some 
people return errors if you return a non-changeable value that is affected by 
a changeable value which the Initiator has modified. Not all the interactions 
of fields are apparent to the host, and no rules govern how each manufacturer 
handles these interactions. It would be much better if the normal action was 
to allow non-changeable fields to be ignored by the target. 
 
If this action appears unsuitable I would recommend the following change be 
done on page 7-31 of revision 10c.
 
"The target shall terminate the MODE SELECT command with CHECK CONDITION 
status, set the sense key to RECOVERED ERROR and set the additional sense code 
to ROUNDED PARAMETER for the following condition:
1) If the Initiator attempts to change any field that is not changeable as 
reported by the target. In this case, no non-changeable parameters shall be 
changed by this command."
"The target shall terminate the MODE SELECT command with CHECK CONDITION 
status, set the sense key to ILLEGAL REQUEST and set the additional sense code 
to INVALID FIELD IN PARAMETER LIST for the following conditions:
1) If the Initiator attempts to send an unsupported value...."