FW: SCAM selection

Tak Asami (Asami asami at dt.wdc.com
Wed Jul 13 15:30:40 PDT 1994

Gary Dickson of Q-Logic says:
> SCAM selection has a timeout of 250 ms, and no the SCAM master does not
> have to play out all of the SCAM cyles before figuring out that no one is
> out there.
> The device that initiates SCAM selection:
> 1. Arbitrates for the bus (and wins the the arbitration).
> 2. Aserts SEL and MSG.
> 3. Deasserts BSY.
> 4. Waits a minimun of a SCAM selection time (250 ms).
> 5. Deasserts MSG.
> 6. Waits for MSG to be deasseted by all (if any) other devices
>    that responded to SCAM selection.
> 7. The SCAM master asserts BSY followed by I/O, C/D, and DB6.  The master
>    then deasserts SEL
> 8. SCAM devices that require soft ID assignement assert BSY followed by I/O and DB7.
>    These devices now deassert SEL.
> 9. After SEL has been released by all devices then by inspecting the bus the following
>    can be determined:
>     1.  If C/D is not asserted there is no SCAM master device. End of SCAM dance
>     2.  If DB7 is not asserted there are no SCAM slave devices requiring ID assignment.
>         End of SCAM dance.
> 10.  If you get this far, there are SCAM slaves that need an ID and there is a SCAM master
>      device.  Let the SCAM continue...

I think he is not correct. The scenario does not seem to conform to the
SCAM protocol described in SPI rev 12c.

According to that version, both the master and slave assert DB6 and DB7, 
so you can't tell that no one responded by looking at DB7.

Norm Harris of Adaptec says:
> At least one iteration will occur in which the SCAM master will determine 
> a null length transfer cycle..i.e. the very first transfer cycle will 
> indicate that no one is driving data.  He will conclude from that that 
> no SCAM devices are present.  The opposite is true for a target..
> there is a mechanism namely if C/D is driven during the SCAM selection, 
> a master is present.

This sounds right. At least this is how we are implementing it....
Do we have a consensus on this, or is there still yet another answer??

== In case you missed, here's my original question... ================
> > 
> > Hi there, SPI fans, this one's for you....
> > 
> > How does a SCAM master device initiating a SCAM selection determine that
> > no other device has responded to the selection attempt?  Is there such a
> > thing as a time-out of a SCAM selection attempt or does the SCAM master
> > figure out no one else is out there from the subsequent SCAM transfer
> > cycles (i.e., he plays out the transfer cycles all by himself before
> > deciding there's no one else out there)?
> > 
> > I thought I understood the process until someone asked me.
> > 

Tak Asami ===============================================================
Western Digital Corp
I/O Product Engineering
asami at dt.wdc.com

More information about the T10 mailing list