LVD SCSI (start round 2)

George Penokie GOP at rchvmp3.vnet.ibm.COM
Thu Oct 17 15:34:51 PDT 1996


* From the SCSI Reflector (scsi at symbios.com), posted by:
* "George Penokie" <GOP at RCHVMP3.VNET.IBM.COM>
*
LVD SCSI (start round 2)

First I want to make clear (which may not be possible) what I do NOT have
any problem with in the DIFFSENS area and why.

-I have no issues with power on or reset.  At both POR and reset I would
always automatically set my device up to run narrow/async and would attempt
to negotiate wider/faster at some later time.

-I have no issues with the bebounce time that is currently described in
SPI-2 on page 22 that states the DIFFSENS shall remain stable for at least
100 ms before it is considered changed to a new level.  BUT I do have a new
concern based on Paul Aloisi's comment:

>Hot plugging SE onto a multimode bus is nasty to say the least. The
>switch over to SE by all the devices is in 100s of milliseconds to
>prevent ground noise from ever triggering a change. A reset or 500
>millisecond delay should be implemented, then unless it is a very
>short bus - <0.5 meters Fast-20 is the maximum speed, but in reality
>on a long bus this may be async is the highest speed that can be
>run on the bus.

I interrupt this as saying the 100 ms minimum debounce time in SPI-2 is not
long enough.  If that is the case then there are more problems than I
thought.

Now back to the issue I have.

Paul also stated in his note:

>If the bus is configuration by be to the LVD rules which may mean
>that Fast-20 SE may not work. Error checking with backoff speeds need
>to be worked into the controllers. The user should be flagged that
>the bus has changed to SE.

Yes I agree something must happen but what???  The standard should tell me
this it should not be left up to the imagination of the implementors as to
what to do in this case.  It is very likely some implementors will not even
notice something needs to be done.

Another comment that was made was that SPI-2 only supports mode 1 therefore
the condition I described was illegal.  WRONG. SPI-2 on page 44 states that
only mode 4 is not supported.  This means I can quiesce the bus then insert
a SE device on a LVD bus and then end up with the problems I stated in my
other note.

I believe wording must be added to SPI-2 to warn this can happen something
like I put in my original note.

> I believed that SPI-2 should make it clear that the DIFFSENS line
> can change at any time and the devices hardware and software must
> immediately respond to any of those changes.
>

I also think the ideas of Tony Priborsky need to be placed into SPI-2.

> After thinking about it a bit, it became clear to me that device (and
> probably host adaptor) firmware must actively monitor the state of the
> DIFFSENS signal and react to changes in it.  SPI-2 does specify some
> debounce intervals, but doesn't say what to do if it changes after initial
> power up.  I was going to suggest that one way to handle it would be to
> treat the event as a kind of SCSI bus reset, and have the devices affected
> process it as such (i.e., reverting to 8-bit, asynch mode).  A new Unit
> Attention ASC/ASCQ pair would indicate that the bus configuration had
> changed.
>

However, the best solution I received was from Wendell Lengefeld:

> I agree that devices and system adapters must immediately respond or act
> upon any change in the DIFFSENS. I would also recommend that we go a
> step further and require that synchronous and wide negotiations be
> step further and require that synchronous and wide negotiations be
> initiated after such a change and would further stipulate that in SE mode
> all devices and adapters must negotiate for no more than Fast 20 speed.
> Fast 40 or greater should only be negotiated in an LVDS environment.

In any case the communications over the last few days indicate this needs
to be discussed in the next working group.  John are you listening???

Bye for now,
George













*
* For SCSI Reflector information, send a message with
* 'info scsi' (no quotes) in the message body to majordomo at symbios.com




More information about the T10 mailing list