arcnet-hardware.txt

《嵌入式系统设计与实例开发实验教材二源码》Linux内核移植与编译实验

 
-----------------------------------------------------------------------------
1) This file is a supplement to arcnet.txt.  Please read that for general
   driver configuration help.
-----------------------------------------------------------------------------
2) This file is no longer Linux-specific.  It should probably be moved out of
   the kernel sources.  Ideas?
-----------------------------------------------------------------------------

Because so many people (myself included) seem to have obtained ARCnet cards
without manuals, this file contains a quick introduction to ARCnet hardware,
some cabling tips, and a listing of all jumper settings I can find. Please
e-mail [e-mail] with any settings for your particular card,
or any other information you have!


INTRODUCTION TO ARCNET
----------------------

ARCnet is a network type which works in a way similar to popular Ethernet
networks but which is also different in some very important ways.

First of all, you can get ARCnet cards in at least two speeds: 2.5 Mbps
(slower than Ethernet) and 100 Mbps (faster than normal Ethernet).  In fact,
there are others as well, but these are less common.  The different hardware
types, as far as I'm aware, are not compatible and so you cannot wire a
100 Mbps card to a 2.5 Mbps card, and so on.  From what I hear, my driver does
work with 100 Mbps cards, but I haven't been able to verify this myself,
since I only have the 2.5 Mbps variety.  It is probably not going to saturate
your 100 Mbps card.  Stop complaining. :)

You also cannot connect an ARCnet card to any kind of Ethernet card and
expect it to work.  

There are two "types" of ARCnet - STAR topology and BUS topology.  This
refers to how the cards are meant to be wired together.  According to most
available documentation, you can only connect STAR cards to STAR cards and
BUS cards to BUS cards.  That makes sense, right?  Well, it's not quite
true; see below under "Cabling."

Once you get past these little stumbling blocks, ARCnet is actually quite a
well-designed standard.  It uses something called "modified token passing"
which makes it completely incompatible with so-called "Token Ring" cards,
but which makes transfers much more reliable than Ethernet does.  In fact,
ARCnet will guarantee that a packet arrives safely at the destination, and
even if it can't possibly be delivered properly (ie. because of a cable
break, or because the destination computer does not exist) it will at least
tell the sender about it.

Because of the carefully defined action of the "token", it will always make
a pass around the "ring" within a maximum length of time.  This makes it
useful for realtime networks.

In addition, all known ARCnet cards have an (almost) identical programming
interface.  This means that with one ARCnet driver you can support any
card, whereas with Ethernet each manufacturer uses what is sometimes a
completely different programming interface, leading to a lot of different,
sometimes very similar, Ethernet drivers.  Of course, always using the same
programming interface also means that when high-performance hardware
facilities like PCI bus mastering DMA appear, it's hard to take advantage of
them.  Let's not go into that.

One thing that makes ARCnet cards difficult to program for, however, is the
limit on their packet sizes; standard ARCnet can only send packets that are
up to 508 bytes in length.  This is smaller than the Internet "bare minimum"
of 576 bytes, let alone the Ethernet MTU of 1500.  To compensate, an extra
level of encapsulation is defined by RFC1201, which I call "packet
splitting," that allows "virtual packets" to grow as large as 64K each,
although they are generally kept down to the Ethernet-style 1500 bytes.

For more information on the advantages and disadvantages (mostly the
advantages) of ARCnet networks, you might try the "ARCnet Trade Association"
WWW page:
	http://www.arcnet.com


CABLING ARCNET NETWORKS
-----------------------

This section was rewritten by 
        Vojtech Pavlik     <[e-mail]>
using information from several people, including:
        Avery Pennraun     <[e-mail]>
 	Stephen A. Wood    <[e-mail]>
 	John Paul Morrison <[e-mail]>
 	Joachim Koenig     <[e-mail]>
and Avery touched it up a bit, at Vojtech's request.

ARCnet (the classic 2.5 Mbps version) can be connected by two different
types of cabling: coax and twisted pair.  The other ARCnet-type networks
(100 Mbps TCNS and 320 kbps - 32 Mbps ARCnet Plus) use different types of
cabling (Type1, Fiber, C1, C4, C5).

For a coax network, you "should" use 93 Ohm RG-62 cable.  But other cables
also work fine, because ARCnet is a very stable network. I personally use 75
Ohm TV antenna cable.

Cards for coax cabling are shipped in two different variants: for BUS and
STAR network topologies.  They are mostly the same.  The only difference
lies in the hybrid chip installed.  BUS cards use high impedance output,
while STAR use low impedance.  Low impedance card (STAR) is electrically
equal to a high impedance one with a terminator installed.

Usually, the ARCnet networks are built up from STAR cards and hubs.  There
are two types of hubs - active and passive.  Passive hubs are small boxes
with four BNC connectors containing four 47 Ohm resistors:

   |         | wires
   R         + junction
-R-+-R-      R 47 Ohm resistors
   R
   |

The shielding is connected together.  Active hubs are much more complicated;
they are powered and contain electronics to amplify the signal and send it
to other segments of the net.  They usually have eight connectors.  Active
hubs come in two variants - dumb and smart.  The dumb variant just
amplifies, but the smart one decodes to digital and encodes back all packets
coming through.  This is much better if you have several hubs in the net,
since many dumb active hubs may worsen the signal quality.

And now to the cabling.  What you can connect together:

1. A card to a card.  This is the simplest way of creating a 2-computer
   network.

2. A card to a passive hub.  Remember that all unused connectors on the hub
   must be properly terminated with 93 Ohm (or something else if you don't
   have the right ones) terminators.
   	(Avery's note: oops, I didn't know that.  Mine (TV cable) works
	anyway, though.)

3. A card to an active hub.  Here is no need to terminate the unused
   connectors except some kind of aesthetic feeling.  But, there may not be
   more than eleven active hubs between any two computers.  That of course
   doesn't limit the number of active hubs on the network.
   
4. An active hub to another.

5. An active hub to passive hub.

Remember, that you can not connect two passive hubs together.  The power loss
implied by such a connection is too high for the net to operate reliably.

An example of a typical ARCnet network:

           R                     S - STAR type card              
    S------H--------A-------S    R - Terminator
           |        |            H - Hub                         
           |        |            A - Active hub                  
           |   S----H----S                                       
           S        |                                            
                    |                                            
                    S                                            
                                                                          
The BUS topology is very similar to the one used by Ethernet.  The only
difference is in cable and terminators: they should be 93 Ohm.  Ethernet
uses 50 Ohm impedance. You use T connectors to put the computers on a single
line of cable, the bus. You have to put terminators at both ends of the
cable. A typical BUS ARCnet network looks like:

    RT----T------T------T------T------TR
     B    B      B      B      B      B

  B - BUS type card
  R - Terminator
  T - T connector

But that is not all! The two types can be connected together.  According to
the official documentation the only way of connecting them is using an active
hub:

         A------T------T------TR
         |      B      B      B
     S---H---S
         |
         S

The official docs also state that you can use STAR cards at the ends of
BUS network in place of a BUS card and a terminator:

     S------T------T------S
            B      B

But, according to my own experiments, you can simply hang a BUS type card
anywhere in middle of a cable in a STAR topology network.  And more - you
can use the bus card in place of any star card if you use a terminator. Then
you can build very complicated networks fulfilling all your needs!  An
example:

                                  S
                                  |
           RT------T-------T------H------S
            B      B       B      |
                                  |       R
    S------A------T-------T-------A-------H------TR                    
           |      B       B       |       |      B                         
           |   S                 BT       |                                 
           |   |                  |  S----A-----S
    S------H---A----S             |       | 
           |   |      S------T----H---S   |
           S   S             B    R       S  
                                                               
A basically different cabling scheme is used with Twisted Pair cabling. Each
of the TP cards has two RJ (phone-cord style) connectors.  The cards are
then daisy-chained together using a cable connecting every two neighboring
cards.  The ends are terminated with RJ 93 Ohm terminators which plug into
the empty connectors of cards on the ends of the chain.  An example:

          ___________   ___________
      _R_|_         _|_|_         _|_R_  
     |     |       |     |       |     |      
     |Card |       |Card |       |Card |     
     |_____|       |_____|       |_____|          


There are also hubs for the TP topology.  There is nothing difficult
involved in using them; you just connect a TP chain to a hub on any end or
even at both.  This way you can create almost any network configuration. 
The maximum of 11 hubs between any two computers on the net applies here as
well.  An example:

    RP-------P--------P--------H-----P------P-----PR
                               |
      RP-----H--------P--------H-----P------PR
             |                 |
             PR                PR

    R - RJ Terminator
    P - TP Card
    H - TP Hub

Like any network, ARCnet has a limited cable length.  These are the maximum
cable lengths between two active ends (an active end being an active hub or
a STAR card).

		RG-62       93 Ohm up to 650 m
		RG-59/U     75 Ohm up to 457 m
		RG-11/U     75 Ohm up to 533 m
		IBM Type 1 150 Ohm up to 200 m
		IBM Type 3 100 Ohm up to 100 m

The maximum length of all cables connected to a passive hub is limited to 65
meters for RG-62 cabling; less for others.  You can see that using passive
hubs in a large network is a bad idea. The maximum length of a single "BUS
Trunk" is about 300 meters for RG-62. The maximum distance between the two
most distant points of the net is limited to 3000 meters. The maximum length
of a TP cable between two cards/hubs is 650 meters.


SETTING THE JUMPERS
-------------------

All ARCnet cards should have a total of four or five different settings:

  - the I/O address:  this is the "port" your ARCnet card is on.  Probed
    values in the Linux ARCnet driver are only from 0x200 through 0x3F0. (If
    your card has additional ones, which is possible, please tell me.) This
    should not be the same as any other device on your system.  According to
    a doc I got from Novell, MS Windows prefers values of 0x300 or more,
    eating net connections on my system (at least) otherwise.  My guess is
    this may be because, if your card is at 0x2E0, probing for a serial port
    at 0x2E8 will reset the card and probably mess things up royally.
	- Avery's favourite: 0x300.

  - the IRQ: on  8-bit cards, it might be 2 (9), 3, 4, 5, or 7.
             on 16-bit cards, it might be 2 (9), 3, 4, 5, 7, or 10-15.
             
    Make sure this is different from any other card on your system.  Note
    that IRQ2 is the same as IRQ9, as far as Linux is concerned.  You can
    "cat /proc/interrupts" for a somewhat complete list of which ones are in
    use at any given time.  Here is a list of common usages from Vojtech
    Pavlik <[e-mail]>:
    	("Not on bus" means there is no way for a card to generate this
	interrupt)
	IRQ  0 - Timer 0 (Not on bus)
	IRQ  1 - Keyboard (Not on bus)
	IRQ  2 - IRQ Controller 2 (Not on bus, nor does interrupt the CPU)
	IRQ  3 - COM2
	IRQ  4 - COM1
	IRQ  5 - FREE (LPT2 if you have it; sometimes COM3; maybe PLIP)
	IRQ  6 - Floppy disk controller
	IRQ  7 - FREE (LPT1 if you don't use the polling driver; PLIP) 
	IRQ  8 - Realtime Clock Interrupt (Not on bus)
	IRQ  9 - FREE (VGA vertical sync interrupt if enabled)
	IRQ 10 - FREE
	IRQ 11 - FREE
	IRQ 12 - FREE
	IRQ 13 - Numeric Coprocessor (Not on bus)
	IRQ 14 - Fixed Disk Controller
	IRQ 15 - FREE (Fixed Disk Controller 2 if you have it) 
	
	Note: IRQ 9 is used on some video cards for the "vertical retrace"
	interrupt.  This interrupt would have been handy for things like
	video games, as it occurs exactly once per screen refresh, but
	unfortunately IBM cancelled this feature starting with the original
	VGA and thus many VGA/SVGA cards do not support it.  For this
	reason, no modern software uses this interrupt and it can almost
	always be safely disabled, if your video card supports it at all.
	
	If your card for some reason CANNOT disable this IRQ (usually there
	is a jumper), one solution would be to clip the printed circuit
	contact on the board: it's the fourth contact from the left on the
	back side.  I take no responsibility if you try this.

	- Avery's favourite: IRQ2 (actually IRQ9).  Watch that VGA, though.

  - the memory address:  Unlike most cards, ARCnets use "shared memory" for
    copying buffers around.  Make SURE it doesn't conflict with any other
    used memory in your system!
	A0000		- VGA graphics memory (ok if you don't have VGA)
        B0000		- Monochrome text mode
        C0000		\  One of these is your VGA BIOS - usually C0000.
        E0000		/
        F0000		- System BIOS

    Anything less than 0xA0000 is, well, a BAD idea since it isn't above
    640k.
	- Avery's favourite: 0xD0000

  - the station address:  Every ARCnet card has its own "unique" network
    address from 0 to 255.  Unlike Ethernet, you can set this address
    yourself with a jumper or switch (or on some cards, with special
    software).  Since it's only 8 bits, you can only have 254 ARCnet cards
    on a network.  DON'T use 0 or 255, since these are reserved (although
    neat stuff will probably happen if you DO use them).  By the way, if you
    haven't already guessed, don't set this the same as any other ARCnet on
    your network!
	- Avery's favourite:  3 and 4.  Not that it matters.

  - There may be ETS1 and ETS2 settings.  These may or may not make a
    difference on your card (many manuals call them "reserved"), but are
    used to change the delays used when powering up a computer on the
    network.  This is only necessary when wiring VERY long range ARCnet
    networks, on the order of 4km or so; in any case, the only real
    requirement here is that all cards on the network with ETS1 and ETS2
    jumpers have them in the same position.  Chris Hindy <[e-mail]>
    sent in a chart with actual values for this:
	ET1	ET2	Response Time	Reconfiguration Time
	---	---	-------------	--------------------
	open	open	74.7us		840us
	open	closed	283.4us		1680us
	closed	open	561.8us		1680us
	closed	closed	1118.6us	1680us
    
    Make sure you set ETS1 and ETS2 to the SAME VALUE for all cards on your
    network.
    
Also, on many cards (not mine, though) there are red and green LED's. 
Vojtech Pavlik <[e-mail]> tells me this is what they mean:
	GREEN           RED             Status
	-----		---		------
	OFF             OFF             Power off
	OFF             Short flashes   Cabling problems (broken cable or not
					  terminated)
	OFF (short)     ON              Card init
	ON              ON              Normal state - everything OK, nothing