Response to SPI-4 proposal 01-251

Richard Moore richard.moore at qlogic.com
Thu Aug 30 16:06:04 PDT 2001


* From the T10 Reflector (t10 at t10.org), posted by:
* Richard Moore <richard.moore at qlogic.com>
*
	This is in response to 01-251 which is on the Parallel SCSI Working
Group agenda for September.
	Due to a prior commitment it is unlikely that I will attend that
meeting, so I am responding by way
	of the reflector.

	The proposal adds restrictions to initiator designs. There is
existing silicon that would be broken
	by this proposal. It is allowable in SPI-3, and in SPI-4 as
currently drafted, for an initiator to count
	REQs in DT mode by counting the trailing edges of REQ only (and
multiplying by two). The proposed
	change would make such initiators illegal under SPI-4.

	The proposal claims that this change fixes a deadlock condition.
With the initiator behavior I described,
	such a deadlock could only occur if a target sent a REQ assertion,
and then ceased to generate REQ
	transitions even when the initiator has ACKed all data prior to the
REQ assertion. In this case, the offset
	will be 1 when the "deadlock" is reached. But unless the negotiated
maximum offset is also 1 (which
	is clearly a useless implementation of synchronous transfers), there
is no deadlock unless the target
	fails to take advantage of the available offset by issuing more
REQs. For a target to wait for zero offset
	is contrary to the purpose of synchronous transfers because it
imposes a performance hit. In order to
	achieve maximum transfer rates, it is up to the target to generate
as many REQs as possible up to the
	maximum offset (subject to other protocol constraints, of course).

	There are no protocol constraints that would force a target to stop
REQing indefinitely with an offset of 1.
	Is there any valid reason to expect that a target would do so?

	Richard Moore
	QLogic Corporation
*
* 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