TST bit logic error in SAM-2 specification
Mark Brown
mkbrown_san at hotmail.com
Fri Apr 19 14:32:20 PDT 2002
* From the T10 Reflector (t10 at t10.org), posted by:
* "Mark Brown" <mkbrown_san at hotmail.com>
*
"It is too difficult to implement (i.e., takes too much code space to handle
both cases) both methods in a target and allow an initiator to choose how it
is done."
I have to disagree here: I believe that the added ability to maintain
separate Mode Pages adds an enormous amount of code to the code base alone.
In addition, all the code that would be required to ensure that initiators
did not cause inconsistencies with the settings of others (I would assume
that the Mode Pages specific to specific devices like the Format Device Mode
Page or the Notch and Partition Mode Page would be off limits). However,
virtualizing the retrieval of basic operational settings out of the Target
Device information blocks into the various Initiator information blocks
constitutes significant software development effort.
However, from a quick look at ACLs I would assume that code to implement
Proxy LUNs for device management administrator access would have to
implement some duplication of Mode Pages for target devices and the
Changeable settings thereof, so this Mode Page virtualization code may have
to be implemented anyway for those who want remote management functionality.
All this wanders from the basic issue which is that TST has definite logic
and state significance (which originates from the need to store Sense error
information indestructibly on a per initiator basis allowing other
initiators to run) within the Device Server and Task Manager and the logic.
The Task Manager should not be making logic decisions about task management
which involve specific initiators' task queues while TST = 000b.
So, to reiterate the central point:
the logic stated in SAM-2 5.8.1.2 Establishing a CA or ACA, clause QErr=11b
and TST=000b is wrong and antithetical to the purpose of the TST bit.
Mark Brown, Micronet Software
>From: Gerry.Houlder at seagate.com
>To: t10 at t10.org
>Subject: RE: TST bit logic error in SAM-2 specification
>Date: Fri, 19 Apr 2002 14:04:27 -0500
>
>* From the T10 Reflector (t10 at t10.org), posted by:
>* Gerry.Houlder at seagate.com
>*
>
> To para-phrase Rob's response, the standard requires the TST field to be
>implemented as "one copy shared by all initiators" even if the target
>implements separate copies of other mode parameters. Even on an
>implementation that does separate copies of mode pages for each initiator,
>this parameter and a number of other parameters must be implemented as
>"shared by all initiators".
>
>A target (or more correctly, a LUN) will not change the way it saves mode
>pages based on this mode setting or any other mode page setting. It is too
>difficult to implement (i.e., takes too much code space to handle both
>cases) both methods in a target and allow an initiator to choose how it is
>done. I would never assume a target will change this characteristic of its
>behavior based on a mode page setting -- the choice will be based on a
>compile time option.
>
>In summary: the system controls this target behavior (if it cares) by
>specifying the mode page behavior in a purchase specification, not by a
>mode page setting. Most systems are designed to tolerate the "one copy
>shared by all initiators" case and this will also tolerate the "copy per
>initiator" case also.
>
>The wording of the sense code is "MODE PARAMETERS CHANGED" because there
>might be only one shared parameter changed, multiple shared parameters
>changed, or even could mean a MODE SELECT was done that might have changed
>a mode setting (in this case there might not have been an actual mode
>parameter change anywhere). If the initiator cares, it has to check all the
>mode pages it cares about to see if something has changed. All of these
>cases (not sure - could be zero, one, more than one) will report the same
>ASC/ASCQ.
>
>
>
>
> "Mark Brown"
> <mkbrown_san at ho To: t10 at t10.org
> tmail.com> cc:
> Sent by: Subject: RE: TST bit logic
>error in SAM-2
> owner-t10 at t10.o specification
> rg
>
>
> 04/19/02 11:43
> AM
>
>
>
>
>
>
>* From the T10 Reflector (t10 at t10.org), posted by:
>* "Mark Brown" <mkbrown_san at hotmail.com>
>*
>Excuse me I did not see this message until just now. Thank you for
>responding. My response:
>
>The implication of the Unit Attention sent to all initiators is that some
>major operational change has occurred at the Target. This implies that the
>TST bit enables and disables the maintenance of Mode Page state information
>on a per initiator basis.
>This further indicates that, as I have suggested, the TST bit has more
>power
>than simply to enable multiple queues for Abort Task Set and Clear Task Set
>purposes, but actually has been given broad operational powers enabling the
>maintenance of State information by the Device Server on a per initiator
>basis (both Request Sense information and now Mode Page information).
>
>Let's put it this way: I do not believe that a functionality as pervasive
>in effect inside the Device Server as well as the Task Manager as
>maintaining separate Mode Pages per initiator could be specified without an
>explicit ability to Enable and Disable it, let alone to interrogate its
>current setting. Therefore, it is fair to assume that that portion of the
>SPC-3 8.4.6 text is stating that the TST bit enables/disables multiple mode
>pages.
>If TST does not enable/disable multiple mode pages, some other bit should.
>
>If the intent is the opposite then at the least the text should read:
>"...shall reflect in all initiator mode pages the new state of the TST
>field selected by...", or
>"...the state of the TST field, if changeable, shall reflect in all
>initiator mode pages the selection by..."
>
>The plural ASC "MODE PARAMETERS CHANGED" also implies more than just the
>TST
>field has changed,
>again indicating that the last initiator's whole Control page (containing
>the TST field) just became the Control page for all initiators. And that
>subsequent changes to other pages by the same or other initiators would
>have
>the same effect on the other initiators.
>
>Mark Brown, Micronet Software
>
> >From: "Elliott, Robert" <Robert.Elliott at COMPAQ.com>
> >To: <t10 at t10.org>
> >Subject: RE: TST bit logic error in SAM-2 specification
> >Date: Wed, 17 Apr 2002 17:56:49 -0500
> >
> >* From the T10 Reflector (t10 at t10.org), posted by:
> >* "Elliott, Robert" <Robert.Elliott at COMPAQ.com>
> >*
> > > -----Original Message-----
> > > From: Mark Brown [mailto:mkbrown_san at hotmail.com]
> > > Sent: Wednesday, April 17, 2002 4:15 PM
> > > Subject: TST bit logic error in SAM-2 specification
> > >
> > > ...
> > > Furthermore, SPC-3 8.4.6 implies that Mode Pages may be maintained
> > > separately for each initiator when TST = 001b, though this
> > > portion of the text is ambiguous as to whether TST requires
> > > and/or enables this functionality.
> >
> >The reference to mode pages in 8.4.6 means there is only one
> >TST value per logical unit - it is shared - regardless of
> >whether the other fields in the Control mode page are shared.
> >If the Control mode page is not shared and one initiator
> >changes the TST field, all the others get a unit attention.
> >This is not true of the other fields.
> >
> >This was one of the reasons the text was reworded in revision 4;
> >see 02-016r1.
> >
> >---
> >Rob Elliott, Compaq? Server Storage
> >Robert.Elliott at compaq.com
> >
> >*
> >* For T10 Reflector information, send a message with
> >* 'info t10' (no quotes) in the message body to majordomo at t10.org
>
>
>_________________________________________________________________
>Chat with friends online, try MSN Messenger: http://messenger.msn.com
>
>*
>* For T10 Reflector information, send a message with
>* 'info t10' (no quotes) in the message body to majordomo at t10.org
>
>
>
>*
>* For T10 Reflector information, send a message with
>* 'info t10' (no quotes) in the message body to majordomo at t10.org
_________________________________________________________________
Join the worlds largest e-mail service with MSN Hotmail.
http://www.hotmail.com
*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org
More information about the T10
mailing list