Support for the new 2940U2W

[Doug Ledford]

On 20-Jan-98 Justin T. Gibbs wrote:
>Actually there are other changes in the 789[01] that require "not
>so minor" driver changes.  I started looking into this over the
>weekend and have made some initial design decisions on how to deal
>with these chips and I hope to have support in the CAM driver later
>this week.  I have a 2940U2W here to test the changes on, but I
>don't have any U2 devices yet.

Ohhh, great.  Don't tell me they did like they did with the 7895 and just
made some registers "go away" :(  Ughh.  For those that can't tell, no, I
haven't gotten my materials on the 789[0,1] chips yet, although Adaptec is
suppossed to be sending me some (as well as my materials on the 7895 that I
don't have yet either).  Well, in that case, if the sequencer doesn't change
*too* much (hint, hint :) then it shouldn't be that hard to add the support
to my patch as well by including the sequencer from Justin and making the
appropriate kernel mods as well.

However, of note to people out there that may be reading this, according to
the Adaptec web site anyway, the 789[01] chips only support the Ultra2
speeds when being used on their Low-Voltage Differential bus.  Additionally,
you can't plug normal SCSI devices directly into the LVD bus.  So, they've
made a chip that converts from LVD to normal Single Ended SCSI bus
semantics, and are selling that separately from the main chip itself. 
People designing cards or motherboards that use this chip would be wise
(IMHO) to get both chips and provide two bus connectors, one LVD and one
Single Ended.  Then, the only requirement between the two busses is that
SCSI ID numbers don't clash between devices, and the Single Ended SCSI bus
will only be good for at most 40MB/s (assuming 20MHz Wide transfers) while
the Ultra2 bus will do twice that fast.  Wihtout something like this, if you
plug so much as one normal Single Ended SCSI device onto that Ultra2 bus,
then no other device on that bus will be able to operate at Ultra2 speeds. 
The reason for this (as from the web site anyway) is that the bus drivers in
that chipset will autodetect any single ended device on that bus, if found,
the bus is driven as a single ended bus instead of a LVD bus, and the max
speed is then capped at the current Ultra Wide speed instead of Ultra2.  If
I remember correctly, the SCSI devices that are Ultra2 compliant are
suppossed to do this exact same thing, if they detect any single ended
devices on the SCSI chain, then they are suppossed to put their own
electronics into single ended mode, not LVD mode, and run that way.

Now, I know why they want LVD for Ultra2 speeds (it's more reliable and
robust in how it does the signalling, just like differential is now, for
instance, it can handle longer cable lengths and what not, plus with the
lower voltage there is less signal noise generated, etc), what I don't know
is if someone writing their own sequencer can "cheat" on this and still run
an Ultra2 device at Ultra2 speeds regardless of the bus type in use.  This
would depend on if the hardware is the one that implements the speed cap
under single ended bus conditions, or if the Adaptec supplied software
drivers are the one that implements the cap.  If it's software, we may be
able to say "We don't care" and go for the unreliable transfer :)  Do you
know on this one Justin?  I would assume that's in the databook somewhere.

----------------------------------
E-Mail: Doug Ledford <dledford at dialnet.net>
Date: 20-Jan-98
Time: 05:47:28
----------------------------------