Re: XAUI and 64b/66b
Howard,
I feel very comfortable with an XGMII that defines IDLES in a way
similar to the one used in the MII and GMII or in the way
you defined in your "10Gig MII update" presentation.
The replacement of the TX_EN and RX_DV signals by four
control signals in your proposal is only a self- evident extension
since we want to interchange now four octets per clock cycle
instead of only one. The embedded delimiters SOP, EOP, ERROR,
enter into the same category of self-evident extensions of
the MII-GMII case to the 10 GbE case.
In the MII and GMII the MAC and PCS/SERDES ignore
what is signaled at TXD[7:0] and RXD[7:0] when TX_EN
and RX_DV are deasserted, respectively. In other words,
the MAC does not send/receive relevant information during
the "IDLE" periods.
What Rich Taborek proposes now is to use the state of
deassertion of the TX_EN or RX_DV to interchange new
codes (K28.0, K28.1, K28.2, K28.3, K28.4, K28.5, K28.6,
K28.7, K23.7, nine total meanwhile). With this addition,
if would be a stretch of the imagination to call this interface
"XGMII". This is not MII or GMII just "10 times faster"
(or 2.5 times faster, since we send four octets every clock
period).
Whether this new proposed interface is better than the
"10 times faster GMII" is something that deserves to
be discussed, as any new proposal does.
Jaime
Jaime E. Kardontchik
Micro Linear
San Jose, CA 95131
Howard Frazier wrote:
> Actually, the MII and GMII do "speak IDLES". Please
> refer to 22.2.3.1 (for the MII) and 35.2.3.1 (for the GMII),
> which describe the "inter-frame" period. 22.2.3.1 is reproduced
> in its entirety below:
>
> 22.2.3.1 Inter-frame
>
> The inter-frame period provides an observation window for an
> unspecified amount of time during which no data activity occurs
> on the MII. The absence of data activity is indicated by the
> de-assertion of the RX_DV signal on the receive path, and the
> de-assertion of the TX_EN signal on the transmit path. The MAC
> interFrameSpacing parameter defined in Clause 4 is measured
> from the de-assertion of the CRS signal to the assertion of
> the CRS signal.
>
> For the XGMII, we have proposed using "embedded delimiters", rather
> than discrete signals like RX_DV and TX_EN to indicate the transition
> between the inter-frame period and the preamble/sfd/data/efd phases.
> That is why we refer to an XGMII code point named "IDLE". This
> code point is nothing more than the way the inter-frame period is
> conveyed across the XGMII, in the absence of discrete signals.
>
> Howard Frazier
> Cisco Systems, Inc.