Subject: Re: Sbus ethernet cards
To: David Laight <port-sparc@netbsd.org>
From: Don Yuniskis <auryn@gci-net.com>
List: port-sparc
Date: 11/26/2001 14:04:35
>> > There are single slot sbus NICs out there aplenty; I do not know if
>> > there are ones with 10BaseT ports, but I own one with AUI and
>> > 10Base2.
>> 
>> I have seen a number of single-slot SBus cards that provide Ethernet.
>
>10M cards - typically an AMD lance tied to the LSI logic Sbus 'DMA' part.

Ah!  That explains the function of that particular QFP!  I had
noticed it on many cards -- even third party cards.  I guess it
is similar to the funky "glue logic" chips for the QBus machines...

>You will see Sun, ICL and Fujitsu parts lurking.
>It ought to be easy to make any of them work.
>ICL certainly made cards that are switchable AUI/10base2, switchable
>AUI/TP and TP only.

But, no AUI/coax/rj45 cards?  (makes life easy for folks like
me who can't make up our minds! :>)

>These cards sometimes contain SCSI because the lump of silicon 
>that interfaces to the Sbus contains a SCSI controller (and a 
>parallel port and...).  There is one on the motherboard that 
>handles a load of the system peripherals.

Oh, OK.  SImilar to a "PC combo chip" (printer, UARTs, RTC, FDC,
IDE, etc.) except the "Sun" part also handles bus control (DMAC)

>(Yes you have Sbus SCSI cards to get more disks, and systems with 2 Sbus
>busses each with 5 empty slots so yuo can add add lots of everything)   


Huh?  Oh, the SBUS connectors are (literally) on a single bus
(perhaps with address decoders mapping each slot to a specific 
address space?)  And, there are machines that have *two* such busses
in them?  Are the controllers for the busses clever enough to
allow activity on both simultaneously?  (e.g., some of the
ARM CPU's let you access DRAM at literally the same instant
that you are accessing ROM/SRAM/IO -- essential to keep
frame buffers alive during slow I/O)

>Known variations:
>
>1) Sun boards do not have a MAC address PROM, 
>ICL (and I think Fj) boards will.

Ouch!  But can you configure a particular hardware address
into the controller thru software?  I.e. tell the NIChip
to ignore the serial eprom interface...?

>2) Later ICL boards (probably) have the 79c90 c-lance 
>because the NMOS part went obsolete - the later motherboards 
>definitely do...

Understood.

>3) It is possible some boards have the DMA+ part - but I 
>didn't ever see any of them.

Presumably this is a driver level issue?  Or, is there a significant
performance win/lose?

>4) The board might have either a pull-up or pull-down on the ALE/AS line
> from the lance.  Both the lance and DMA need to be configured to match
> this resister. Further (from a driver source file):
> /* If the lance is run with ACON set then it glitches ALE when 
> releasing hold.

> They fixed this in the C-Lance - but now get a glitch when ACON is 
> clear.  Thus the Lance must be run in ALE mode and the C-lance in 
> ACON mode
> (failure to do this can cause the DMA+ part to lock the Sbus). */
>
>5) 100M cards based on several sun chipsets, including the FEPS.
>   (I wrote a driver for that one...)


I'm not (currently) looking for 100Mbs cards.  "Take baby steps"  :>

>I don't know how 'good' the netBSD driver for these cards is.  
>We had a lot of trouble with Sbus latency on heavily loaded 
>(usually multi-cpu) systems.

OK.  Yet another issue that I don't have to face (yet!  :>)

>(but never in the development lab....)  The Sbus doesn't 
>guarantee the lance enough bandwith - so you get fifo underrun 
>and overrun errors.


IIRC, the LANCE  only has a small Tx/Rx buffer... less than
100 bytes (maybe 64?)  I.e. no internal "packet memory".

>Yes - these card come up for sale on ebay...


I will have to start "watching" to see what is what...
Thx,
--don