QAS with non-packetized protocol

Richard Moore r_moore at qlc.com
Mon Jun 26 19:02:12 PDT 2000


* From the T10 Reflector (t10 at t10.org), posted by:
* Richard Moore <r_moore at qlc.com>
*
We could require the QAS message to be the first message of a MSG IN phase,
so that it is not necessary to snoop & parse all message bytes -- only the
first byte. But there is a requirement to snoop ATN for the remainder
of the message phase. If ATN is asserted on a subsequent message byte,
the target would change to MSG OUT rather than start the QAS arbitration
phase. Under these circumstances it would be necessary to cancel the
effect of the QAS message, so that the target must either go BUS FREE
following the exchange of messages, or must go back to MSG IN and send
another QAS REQUEST if it still wants to do a QAS.

The only way to break this, I think, is if you have a device implementing
QAS the SPI-3, packetized-only way. Such a device will see the QAS message
and COULD (depending on implementation) ignore the subsequent messages.
(I think SPI-3 does require the snooping device to see ONLY ONE message
in a MSG IN phase in order to detect a valid QAS REQUEST -- not sure
about that, though). Then, if there is an ATN assertion, and the target
switches the MSG signal off briefly during the MSG IN-to-MSG OUT phase
transition, the snooping device could interpret this as the start of the
QAS arbitration. But that failure mode requires several "if"s.

   -- Richard Moore
     QLogic Corp.

* From the T10 Reflector (t10 at t10.org), posted by:
* "Bill Galloway" <BillG at breatech.com>
*
Messages are one of the hardest things in SCSI to get right. It is very
scary
to think that hardware would have to parse and understand messages
correctly.
I can easily come up with 2^7 legal message sequences that may end a
connection.  The real number is closer to 2^9 legal (but strange)
message
sequences. The hardware would have to know and interpret the length of
all
messages to know where message boundaries are.  Currently there are
reserved
messages without any defined length.  We would have to lock down in
hardware
the length of all reserved messages.  If in the future we had a good
reason to
make message codes 0x30-0x3f three bytes long we would not be able to.
Two
spi-5 devices could not implement any messages that could not be snooped
by
today's devices. This breaks the current model where only the two
communicating
devices have to understand the message.

Would this hardware have to deal with all possible message
protocol/parity
errors?? I bet someone will create a tester that will check for these.
When
they find a bug you will be turning silicon not reving firmware.

Just the act of snooping is problematic.  We have said in the past that
if a
device did not successfully snoop the QAS REQUEST message then no real
harm was
done, it just did not participate in that arbitration.  If we require
all
devices to parse all messages to find the QAS message then ALL bytes
must be
snooped successfully.  If a device misses a single message byte it can
mis-interpret all remaining message bytes.

I understand that limiting QAS to packetized causes problems for tapes
but
snooping ALL message bytes is NOT the answer.


Bill Galloway
BREA Technologies, Inc.
P: (281) 530-3063
F: (281) 988-0358
BillG at breatech.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




More information about the T10 mailing list