Re: code-word alignment location
Ben,
A couple of points on your note:
1) Link sync should occur separately for each data direction and should not
require any handshakes. Inhibiting transmit packets during initialization
requires handskakes. My rationale is that 8B/10B can achieve lane sync on as
little as a single comma code-group and link sync after a packet or two. 64B/66B
always transmits 66B frames which are used by the receiver to sync up. A link
should be able to resync on its own recognisance, without handshakes.
2) If link is broke it should be shut off. This is the reason for Remote Fault
and Break Link. These signals may be considered to be handshakes, but at the
management level.
3) A link is in sync when Link Status = OK.
Best Regards,
Rich
--
Ben Brown wrote:
>
> Una,
>
> Una Quinlan wrote:
> >
> > ZZZZ/ZZZZ = KKKK/RRRR or RRRR/KKKK or AAAA/KKKK ....
> > (with exact matching of the 7-bit types)
>
> I would probably write this as
>
> ZZZZ/ZZZZ = IIII/IIII
> (with exact matching of the 7-bit types)
>
> since, in my opinion, the /A/s, /K/s and /R/s should
> be part of XAUI only. Other than that I agree with you
> that this is how a 64b/66b link should/could work. I
> also agree with Rich that there are enough IDLEs, even
> when being rudely interrupted by packets, that alignment
> should be possible without handshakes (or at least that
> assumption should be made at the onset). I don't think
> this discussion is particularly relevant to SLP since
> it uses a different alignment method.
>
> Another consideration is that, until alignment occurs,
> transmit packets could be inhibited, similar to how
> it worked in 1000Base-X with sync_status. In 1000Base-X,
> sync_status was tied into the AN machine to return it
> to the AN_ENABLE state and start sending config words
> or idles (based on whether AN was enabled or not). Since
> we're not using AN, we probably won't have an AN machine
> but we can still use loss of sync to force the transmit
> of IDLE.
>
> This way, there is bound to be at least some amount of
> startup time for a link to send some number of IDLEs in
> both directions (as long as it takes for a link to gain
> SYNC).
>
> Ben
>
> Una Quinlan wrote:
> >
> > Ben,
> >
> > This is exactly the conclusion I came to - that the alignment will
> > be in a different layer this time (compared to comma detect in the
> > PMA).
> >
> > So if the 64b/66b PCS needs to do its own alignment, it sounds like it
> >
> > would be important for the PCS to have a link initialisation
> > sequence (which hopefully could be kept very simple). Because
> > it could be near-impossible to align to 0/1 1/0 sync bits, without
> > also having assistance from specific Types. So to facilitate code
> > alignment (locally and link partner), there will need to be some
> > well-defined codewords to sync to. Perhaps this could be handled in
> > the IPG, by searching for
> >
> > ZZZZ/ZZZZ = KKKK/RRRR or RRRR/KKKK or AAAA/KKKK ....
> > (with exact matching of the 7-bit types)
> >
> > - and then follow-on by checking a few sequential codeword boundaries
> > to verify the framing is correct (further 0/1 or 1/0 checks).
> >
> > This places a requirement on the minimum IPG length. And the PCS
> > would only gain alignment in between packets. But this should be okay.
> >
> > Does anyone see a requirement for the PCS to define link-init
> > codewords (eg break-link + handshake) ?
> >
> > Regards,
> > Una
> >
> > -----Original Message-----
> > From: Brown, Ben [BAY:NHBED:DS48] [mailto:bebrown@xxxxxxxxxxxxxxxxxx]
> > Sent: 10 April 2000 14:57
> > To: stds-802-3-hssg@xxxxxxxx
> > Subject: code-word alignment location
> >
> > Una,
> >
> > I thought I'd throw in a response to your aside using a new thread.
> >
> > > Una Quinlan wrote:
> > >
> > > [ Aside: Does PCS codeword alignment belong in the PMA/WIS or the
> > PCS?].
> > >
> >
> > In 1000Base-X, code-word alignment (to the comma) occurred in the PMA.
> >
> > With a break down (from the Uni-PHY proposal) of PMD - PMA - WIS -
> > PCS,
> > the PMA cannot provide alignment. The WIS needs to align on A1/A2.
> > From
> > this, the WIS provides byte aligned data to the PCS. If the PCS
> > is 64b/66b, this byte alignment is useless and 66b alignment is done
> > to the 1/0 or 0/1 sync bits. The PMA can't participate in any of this
> > because the WIS is optional and both the WIS and PCS require different
> >
> > alignment mechanisms. Implementations can do anything they want but
> > the
> > standard needs to keep these separate.
> >
> > Ben
> >
> -----------------------------------------
> Benjamin Brown
> Router Products Division
> Nortel Networks
> 1 Bedford Farms,
> Kilton Road
> Bedford, NH 03110
> 603-629-3027 - Work
> 603-624-4382 - Fax
> 603-798-4115 - Home
> bebrown@xxxxxxxxxxxxxxxxxx
> -----------------------------------------
-------------------------------------------------------
Richard Taborek Sr. Phone: 408-845-6102
Chief Technology Officer Cell: 408-832-3957
nSerial Corporation Fax: 408-845-6114
2500-5 Augustine Dr. mailto:rtaborek@xxxxxxxxxxx
Santa Clara, CA 95054 http://www.nSerial.com