Inside Surfer.txt

Programming docs for the Silversurfer serial port

This document is freely distributable as long as it is not changed, and one
of the sources www.jschoenfeld.com, www.jschonfeld.de, www.siliconsonic.com
is mentioned. 

last update: March 1st, 2001

The silversurfer serial port uses a 16c550 UART Chip by California
Microdevices (www.calmicro.com). The datasheet is not available on
the internet, if you are interested in timing details, contact me directly,
as Calmicro has discontinued the part, and is no longer supporting developers
with datasheets. The UART is compatible with the common 16c550
implementations, so you may not need the details. Just to be complete:
The full part number is CM16C550PE. CM is Calmicro's prefix, and PE marks the
package (PLCC-44 in this case). 

There is somebody on the internet who claims to have developed the
Silversurfer. He gives technical information without any background, and
claims that the datasheet is available on www.cmd.com. Please do not disturb
the guys from CMD, as this is a storage company, they never made chips like
the 16c550. This is just the proof that the guy is lying - he may do good
software, but nothing else.

The silversurfer is available in two versions. The basic version is meant for
the clock-port of the A1200. The 16c550 has eight registers, and some
multiplexed registers for the Baud Rate generator. The address space for the
clock-port is 16 bytes, and the Silversurfer occupies all of them. The upper
eight registers are mirrors of the lower eight registers. In the following
text, I will only talk about the lower eight addresses, but you must keep in
mind that there may be a mirror. I'm using the term "may be" because you have
the possibiliy to disable one of the two blocks by closing a jumper (see
further down).

The second version is the "limited Edition", which has a different connector,
and a different memory map when connected to the 26-pin expansion port. When
connected to the clock-port (with a cable!), the memory map is the same as
with the standard version.

summary of registers:
(note: DLAB is short for Divisor Latch Access Bit, it's bit#7 in register#3)

Number: 0
Address in A1200: $d80001
Address on Limited Edition: Board Offset + port offset + $18

With DLAB=0:
Reading reads the "receiver Buffer Register"
Writing writes to the Transmitter holding register
With DLAB=1:
Divisor Latch (lower significant byte)

Number: 1
Address in A1200: $d80005
Address on Limited Edition: Board Offset + port offset + $1a

With DLAB=0:
interrupt enable register
With DLAB=1:
Divisor Latch (most significant byte)

The following registers are not affected by the DLAB bit:

Number: 2
Address in A1200: $d80009
Address on Limited Edition: Board Offset + port offset + $1c
read: Interrupt identification register
write: Fifo Control register

Number: 3
Address in A1200: $d8000d
Address on Limited Edition: Board Offset + port offset + $1e
Line control register

Number: 4
Address in A1200: $d80011
Address on Limited Edition: Board Offset + port offset + $38
Modem control register

Number: 5
Address in A1200: $d80015
Address on Limited Edition: Board Offset + port offset + $3a
Line status register (read only)

Number: 6
Address in A1200: $d80019
Address on Limited Edition: Board Offset + port offset + $3c
Modem status register

Number: 7
Address in A1200: $d8001d
Address on Limited Edition: Board Offset + port offset + $3e
Scratch pad register
------

With an IRQ enabled, the Silversurfer will issue an IRQ6, no matter where
it is connected. There are no more registers than the chip registers. All
you need is eight addresses.

The Z4 zorro expansion for A1200 computers has four clock-ports. It
is advisable to support them, as many people have bought the Silversurfer
because of the compatibility to this extension:

Add $4000 to all clock-port registers to access a Silversurfer on
clock-port 1 of the Z4 board. 

Add $8000 to all clock-port registers to access a Silversurfer on
clock-port 2 of the Z4 board. 

Add $c000 to all clock-port registers to access a Silversurfer on
clock-port 3 of the Z4 board.

Check for presence of the Z4 board by checking mirrors of the scratch pad
register. If you write to the scratch pad register at $d8001d and also find
that value at $d8401d, there can be no Z4 board. DO CROSS-CHECKS by writing
the scratch pad register in $d8401d and reading it back on $d8001d, as there
can be trash in the scratch pad from the previous run of your (or other)
software. Writes should be done with different bit-patterns in order to
minimize the possiblity of false identification.

Silversurfer on other clock-ports:

The Buddha Flash also contains a clock-port. This port is located at even
addresses, so substract $d80001 from the clock-port values and add the board
offset plus the port offset. In other words, if the Buddha is located at
$ea0000, the eight registers are at:

$ea0e00
$ea0e04
$ea0e08
$ea0e0c
$ea0e10
$ea0e14
$ea0e18
$ea0e1c

Similar calculation with the X-Surf clock-ports. Mind that one of the
ports uses even addresses, and the other uses odd addresses. This has
been done in order to balance the load on the data lines a bit. While
implementing software for other ports, also look at the Inside_ISDN.txt
file for the port location on the ISDN Surfer Zorro board.

The UART is clocked at 7,372800 Mhz. The frequency may sound odd, but
it is perfect for generating the common baud rates. To operate properly,
the Baud rate generator uses a 16x clock rate, so if you want to set 38400
baud, the Baud rate generator must be programmed to genberate a 614400 clock.
This is done by calculating 7372800/614400=12. Write this value to the
Divisor Latch LSB, and a 0 to the Divisor Latch MSB, and you're done.

Solder-jumpers on the back of the silversurfer:

There are six unused pads on the back of the board. Four of them are
maent for pullup/pulldown resistors or noise reduction capacitors. This
relates to my experience with the non-working Hypercom 1 models, that could
have been debugged with 33pF capacitors on IOR and IOW (unfortunately,
somebody decided to use standard low-cost capacitors, so the bug still
exists - it will not if you use X7R capacitors in SMD style!). The chip
used on the Silversurfer does not need this kind of debugging, it has a
built-in input hysteresis, so noise is filtered on-chip. Just leave the
four pads on the Silversurfer unused, and focus on the other two, which
are more interesting: They are simple jumpers, either use a wire or a 0-ohm
resistor to close the jumpers.

Never close both jumpers at the same time!

location of the jumpers:
The diagram shows the silversurfer from the solder-side!

---------------------------
|     clock-port          |
|                         |
|                         |
|                         |
|                         |
------                    -------
     |                          |
     |            1#  #3        |
     |                          |
     |            2#  #4        |
     |                          |
     |                          |
     |                      #  #|
     |                          |
     |     #  #             #  #|
     |                          |
     |     #  #                 |
     |                          |
     |                          |
     |                          |
     |                          |
     |                          |
     |                          |
     |                          |
     |                          |
     |##                        |---
     |##                        |---
     |##                        |---   serial connector
     |##                        |---
     |                          |---
     ----------------------------

Only use pads with numbers, leave the other pads (marked with #) unused!!

Usually, the CS2 signal (pin 16) of the UART is pulled high with a resistor
that is located on the top of the silversurfer. Closing one of the jumpers
applies address line A5 to this pin, either inverted, or double-inverted
(that is, directly). As a result, one of the register banks will be disabled:

Connection 1-2 (R2) will disable the lower register bank
This means, that the UART chip is only located at addresses $d80021 and up
(Mirror on $d80001 and up is free, can be used for other hardware).

Connection 3-4 (R4) will disable the upper register bank
This means, that the UART chip is only located at addresses $d80001 and up
(Mirror on $d80021 and up is free, can be used for other hardware).

These solder-jumpers are also present on the limited edition of the
Silversurfer. Only use them if you are using the clock-port connector,
otherwise you will cause even more confusion with the registers of the
26-pin connector. 

The other unused pads on the opposite side are meant for direct connection of
a DB9-male connector. This has never been used, and will only fit if you
use the Silversurfer on a Zorro board (or in a completely different
enviroment). The orientation of the connector is obvious: Four pins on the
solderside, five pins on the component side will speak for themselves. 

If you have any comments or want to have a hardcopy of the California
Microdevices datasheet, send me an e-mail: jens@jschoenfeld.de
---EOF

This document is freely distributable as long as it is not changed, and one
of the sources www.jschoenfeld.com, www.jschoenfeld.de, www.siliconsonic.com
or www.ami.ga is mentioned. 

last update:
august 31st, 2003: initial release

General
-------
RR-Net is a 10MBit networking card that is designed for the commodore C64,
for use on the Retro Replay freezer cartridge. It'll also work on the Amiga
with a clock-port, but it is not recommended: The card does not support IRQs,
and it hardly fits mechanically. If you need network for your Amiga, use the
X-Surf 2 board.

Memory map
----------
The memory map depends on the place where RR-Net is connected. Although
RR-Net is designed for the C64, here are also the offset addresses for the
Amiga. Check the corresponding hardware documentations for the carrier cards
to obtain the base addresses of the clockport.

RR-Net is based on the Cirrus logic CS8900a (Crystal LAN). The chip is used
in 8-bit mode, so the 8 registers of an NE2000 are spread over 16 registers
in an 8-bit system. The 8-bit mode of the chip does not support IRQs (see
Cirrus logic application note AN181).

Do not forget to switch on the accessory connector of the Retro Replay by
setting bit 0 in $de01. See Inside_Replay.txt for more information.

C64 register    Amiga register  read/write      meaning
none            $00/$04         read-only       interrupt status queue
$de02/$de03     $08/$0c         read/write      PackatPage pointer
$de04/$de05     $10/$14         read/write      PacketPage Data (Port 0)
$de06/$de07     $18/$1c         read/write      PacketPage Data (Port 1)
$de08/$de09     $20/$24         read/write      Receive/Transmit Data (Port 0)
$de0a/$de0b     $28/$2c         read/write      Receive/Transmit Data (Port 1)
$de0c/$de0d     $30/$34         write-only      TxCMD (Transmit Command)
$de0e/$de0f     $38/$3c         write-only      TxLength (Transmit Length)

Although the Amiga makes the interrupt status queue registers available, it
does not have any effect. Even if you try to activate the chip's IRQ features,
it will not have any effect. The IRQ line of the chip is not wired on RR-Net
at all!
If you compare this register map to the register map in AN181, you might
notice that the first and the second half of the register set is swapped.
This has been done, because the accessory connector of the Retro Replay does
not provide the full range of registers available on the Amiga clockport. The
first two bytes are control registers of the freezer cartridge, and this
swapping moves the unnecessary register-pair (IRQ status queue) into the
"invisible" area of the cartridge.

Endianess
---------
Although the chip is used in 8-bit mode, you have to know the endianess,
because each register pair forms a one 16-bit register. Ethernet is
big-endian by default. The chip was designed as an ISA thing, so all
registers appear as little endian. In other words: The lowbyte comes first,
then the highbyte.

MAC address
-----------
RR-Net does not have an EEprom to store the MAC address, so it should be
stored somewhere else: The flash rom of Retro Replay, the disk of the
Contiki operating system by Adam Dunkels (http://dunkels.com/adam/contiki/),
or Kickflash OS4 for the Amiga. The Mac address is only needed when
transmitting packets, so it's OK if it's loaded from mass storage media when
loading/starting the driver. The Mac address is NOT needed when RR-Net is
just connected to other networking equipment. The LINK led might already be
on, but the MAC address is only needed when something is transmitted into the
network.
         
More literature
---------------
Cirrus Logic Application note AN181: "Using the Crystal CS8900A in 8-bit
mode" by James Ayres: http://www.cirrus.com/en/pubs/appNote/an181.pdf

Crystal LAN CS8900A Ethernet Controller Technical Reference manual AN83:
http://www.cirrus.com/en/pubs/appNote/An83-3.pdf

The product data sheet for the CS8900A also contains software information,
starting at chapter 4 (page 38):
http://www.cirrus.com/en/pubs/proDatasheet/cs8900a-4.pdf

Application note 194: "How to program the Hash filter in the CS8900A":
http://www.cirrus.com/en/pubs/appNote/an194-1.pdf

The FAQ about the chip: http://www.cirrus.com/en/pubs/appNote/an205-2.pdf

Fine print
----------
RR-Net is not designed, authorized or warranted to be suitable for use in
life-support devices or systems or other critical operations. Inclusion of
the product in such applications is understood to be fully at the
customer's risk.

** EOF