CLASS="SECT1" BGCOLOR="#FFFFFF" TEXT="#000000" LINK="#0000FF" VLINK="#840084" ALINK="#0000FF" >

3. Which driver should I use?

The realm of Token Ring drivers on Linux has expanded quite a bit in last couple of years. It's not just ibmtr anymore! So as a result this map will tell you given a card which driver you should try and the recommended minimum kernel version (if any).

3COM

IBM

Olicom

For these Olicom cards, see their website http://www.olicom.com for drivers. You will need a 2.2.x series kernel.

Madge

For these Madge cards you'll want to visit their site http://www.madge.com for drivers and get the 2.31 Madge drivers. You will need either a 2.0.36 or 2.2.5 as a minimum.

2.41 drivers:

According to the Madge README file the 2.41 driver has been tested on uniprocessor and SMP kernel versions: 2.0.36, 2.2.5-15 ,2.2.10, 2.2.12-20, 2.4.2-2.

Other Madge cards are reportedly based on the Texas Instruments tms380 chipset and thus as of the 2.3.26 kernel you can try the tms380tr driver.

SysKonnect

In the 2.2.x series of kernels try sktr. In the 2.3.x and greater series try the tms380tr driver.

SMC

Driver is included as part of the 2.3.38+ kernel.

Intel

Support for these cards is currently under development. Check http://www.linuxtr.net for status.

3.1. Drivers/Adapter Specifics

Here we'll describe the different options and configurations available for each of the available drivers.

3.1.1. Kernel Module Aliases and Parameters

Most drivers accept arguments in the form of module paramters (with the exception of the special case of PCMCIA, which is fully described below).

Kernel modules are specified in the file /etc/conf.modules or /etc/modules.conf depending upon which version of modutils you've got.

You can directly modify this file or use the tools builtin to your specific distribution. These distribution specific tools are beyond the scope of this document, but you can always directly modify the modules.conf file by hand to get things up and running and then figure out how your distribution handles these files. For example, Debian has several files in the /etc/modutils directory and from these builds the modules.conf file.

Kernel modules aliases are utilized to associate a particular name with a kernel module.

For token ring, this is used to assign drivers for each of the token ring interfaces so that the system scripts know which driver to insert when you bring an interface up.

The format of the alias lines are:
		alias module_name interface
		
Usually, the only line you'll need for the token ring networking would be something like:
alias olympic tr0
This binds the olympic driver to the tr0 interface so when you type
ifconfig tr0 up
if the tr0 interface is not already loaded, the system will insert the olympic driver, which in turn will find the network card and create the tr0 network device.

Kernel modules parameters are specified in the following format:
		options module_name parameter_1=XXX [parameter2=YYY ...]
		
Where the modules_name is the name of the driver, i.e. olympic, ibmtr, 3c359 and the ` parameters are those available for each driver. See either the following sections for driver specifics or check out the drivers source code.

For example, if you wanted to set the Olympic driver to 16 mbps operation and with a default buffer size of 8192 bytes, you would use the following line:
		options olympic ringspeed=16 pkt_buf_sz=8192
		

3.1.2. IBMTR Driver

IBM Tropic Chipset Based Token Ring Adapters

This is the original token ring driver in the kernel and supports almost all adapters that use the IBM Tropic chipset, including the IBM ISA, ISA/Pnp, and a multitude of adapters from other manufacturers.

The IBM Turbo 16/4 ISA/PnP adapter will, in fact, work fine with the ibmtr driver. In older drivers you had to run the card in Auto 16/4 compatability mode. The simplest way to set this is to use the LANAID disks sent with the card and run the command:
LANAIDC /FAST=AUTO16
You should then use LANAIDC or LANAID to configure the card according to documentation. The latest drivers for the Turbo Adapters will recognize these adapters and configure them straight out of the box. You may have to either turn off isapnp support in the kernel or modify your isapnp.conf file to enable the adapter.

Options:

Perusal of the ibmtr source code may leave you to believe that the adapter can take three parameters, however, in reality the driver doesn't take any. These parameters are a hang over from the early stages of the driver and are only intended to be used to force the driver to only test restricted åddresses when looking for adapters. The information on these options are included here for completeness only.

  • io: Specify the I/O ports that the driver will check for the presence of any cards. All Tropic based ISA adapters, or adapters emulating the ISA cards will be found on either port 0xA20 or 0xA24. If you know that your adapter is configured for 0xA24 and/or that probing on port 0xA20 will cause problems with your machine, use io to force the driver to check a specific port only.

    The Turbo adapters (including the confusingly named latest Auto 16/4 cards) can have their io regions located anywhere permitted by the PnP specification. This location is found using the new turbo detection code and no parameters are required.

  • irq & mem: The two options were used to tell the driver exactly which irq to use and where the shared ram for the adapter could be found. These two options are now totally redundant in the driver as the interrupt line and the location of the shared ram is obtained directly by interrogating the adapter.

3.1.3. Olympic Driver

IBM PCI Pit/Pit-Phy/Olympic chipset based token ring cards

Options:

The driver accepts four options: ringspeed, pkt_buf_sz, message_level and network_monitor.

These options can be specified differently for each card found, i.e if you have two olympic adapters in your machine and want to assign a ring speed of 16mbps to the first adapter, but a ring speed of 4mbps to the second adapter, your options line would read:
		options olympic ringspeed=16,4
		
However, it should be noted that the driver assigns value to each adapter in the order they are discovered¸ which is usually the order there are present on the pci bus. A little trial and error may be required to be certain which adapter is receiving which configuration option.

  • ringspeed: Has one of three settings 0 (default), 4 or 16. 0 will make the card autosense the ringspeed and join at the appropriate speed, this will be the default option for most people. 4 or 16 allow you to explicitly force the card to operate at a certain speed. The card will fail if you try to insert it at the wrong speed. (Although some hubs will allow this so be *very* careful). The main purpose for explicitly setting the ring speed is for when the card is first on the ring. In autosense mode, if the card cannot detect any active monitors on the ring it will not open, so you must re-init the card at the appropriate speed. Unfortunately at present the only way of doing this is rmmod and insmod which is a bit tough if it is compiled in the kernel. The driver does support 100 mbps full duplex operation. This is automatically detected by the adapter when connected to an appropriate switch.

  • pkt_buf_sz: This is this initial receive buffer allocation size. This will default to 4096 if no value is entered. You may increase performance of the driver by setting this to a value larger than the network packet size, although the driver now re-sizes buffers based on MTU settings as well.

  • message_level: Controls level of messages created by the driver. Defaults to 0 which only displays start-up and critical messages. Presently any non-zero value will display all soft messages as well. NB This does not turn debugging messages on, that must be done by modified the source code.

  • network_monitor: Any non-zero value will provide a quasi network monitoring mode. All unexpected MAC frames (beaconing etc.) will be received by the driver and the source and destination addresses printed. Also an entry will be added in /proc/net called olympic_tr%d, where tr%d is the registered device name, i.e tr0, tr1, etc. This displays low level information about the configuration of the ring and the adapter. This feature has been designed for network administrators to assist in the diagnosis of network / ring problems. (This used to OLYMPIC_NETWORK_MONITOR, but has now changed to allow each adapter to be configured differently and to alleviate the necessity to re-compile olympic to turn the option on).

Multi-card. The driver will detect multiple cards and will work with shared interrupts, each card is assigned the next token ring device, i.e. tr0 , tr1, tr2. The driver should also happily reside in the system with other drivers. It has been tested with ibmtr.c running. I have had multiple cards in the same system, all sharing the same interrupt and working perfectly fine together. This is also true for the Cardbus Olympic adapters, I have quite happily had a Cardbus adapter and regular 16 bit PCMCIA token ring adapter working together in the same laptop.

Variable MTU size:. The driver can handle a MTU size upto either 4500 or 18000 depending upon ring speed. The driver also changes the size of the receive buffers as part of the mtu re-sizing, so if you set mtu = 18000, you will need to be able to allocate 16 * (sk_buff with 18000 buffer size) call it 18500 bytes per ring position = 296,000 bytes of memory space, plus of course anything necessary for the tx sk_buff's. Remember this is per card, so if you are building routers, gateway's etc, you could start to use a lot of memory real fast.

3.1.4. Lanstreamer Driver

IBM PCI/MCA Lanstreamer chipset based token ring cards

Options:

The driver accepts three options: ringspeed, pkt_buf_sz, message_level and network_monitor.

These options can be specified differently for each card found, i.e if you have two olympic adapters in your machine and want to assign a ring speed of 16mbps to the first adapter, but a ring speed of 4mbps to the second adapter, your options line would read:
		options lanstreamer ringspeed=16,4
		
However, it should be noted that the driver assigns value to each adapter in the order they are discovered¸ which is usually the order there are present on the pci/mca bus. A little trial and error may be required to be certain which adapter is receiving which configuration option.

  • ringspeed: Has one of three settings 0 (default), 4 or 16. 0 will make the card autosense the ringspeed and join at the appropriate speed, this will be the default option for most people. 4 or 16 allow you to explicitly force the card to operate at a certain speed. The card will fail if you try to insert it at the wrong speed. (Although some hubs will allow this so be *very* careful). The main purpose for explicitly setting the ring speed is for when the card is first on the ring. In autosense mode, if the card cannot detect any active monitors on the ring it will not open, so you must re-init the card at the appropriate speed. Unfortunately at present the only way of doing this is rmmod and insmod which is a bit tough if it is compiled in the kernel. switch.

  • pkt_buf_sz: This is this initial receive buffer allocation size. This will default to 4096 if no value is entered. You may increase performance of the driver by setting this to a value larger than the network packet size, although the driver now re-sizes buffers based on MTU settings as well.

  • message_level: Controls level of messages created by the driver. Defaults to 0 which only displays start-up and critical messages. Presently any non-zero value will display all soft messages as well. NB This does not turn debugging messages on, that must be done by modified the source code.

Network Monitor. The Lanstreamer driver does support a network monitor mode similar to the olympic driver, however it is a compile time option and not a module parameter. To enable the network monitor mode, edit lanstreamer.c and change the line:
#define STREAMER_NETWORK_MONITOR 0
to read:
#define STREAMER_NETWORK_MONITOR 1
All unexpected MAC frames (beaconing etc.) will be received by the driver and the source and destination addresses printed. Also an entry will be added in /proc/net called streamer_tr. This displays low level information about the configuration of the ring and the adapter. This feature has been designed for network administrators to assist in the diagnosis of network / ring problems.

Multi-card. The driver will detect multiple cards and will work with shared interrupts, each card is assigned the next token ring device, i.e. tr0 , tr1, tr2. The driver should also happily reside in the system with other drivers.

Variable MTU size:. The driver can handle a MTU size upto either 4500 or 18000 depending upon ring speed. The driver also changes the size of the receive buffers as part of the mtu re-sizing, so if you set mtu = 18000, you will need to be able to allocate 16 * (sk_buff with 18000 buffer size) call it 18500 bytes per ring position = 296,000 bytes of memory space, plus of course anything necessary for the tx sk_buff's. Remember this is per card, so if you are building routers, gateway's etc, you could start to use a lot of memory real fast.

3.1.5. 3Com 3C359 Driver

3COM PCI TOKEN LINK VELOCITY XL TOKEN RING CARDS

Currently the 3c359 driver in not included in the standard kernel source. To utlize the driver, you must download the driver from the Linux Token Ring Project web site and patch your kernel.

Once you've downloaded the file, you can patch your kernel with the following commands:
		cd /usr/src/linux
		patch -p1 < 3c359-2.4.16.patch
		
or, if the patch file is gzipped:
		zcat 3c359-2.4.16.patch | patch -p1
		
Then just run make config|menuconfig|xconfig and select the 3c359 driver from the token ring drivers section of the kernel configuration and then compile and install the kernel and/or modules as usual.

Options:

The driver accepts three options: ringspeed, pkt_buf_sz, message_level.

These options can be specified differently for each card found, i.e if you have two olympic adapters in your machine and want to assign a ring speed of 16mbps to the first adapter, but a ring speed of 4mbps to the second adapter, your options line would read:
		options 3c359 ringspeed=16,4
		
However, it should be noted that the driver assigns value to each adapter in the order they are discovered¸ which is usually the order there are present on the pci bus. A little trial and error may be required to be certain which adapter is receiving which configuration option.

  • ringspeed: Has one of three settings 0 (default), 4 or 16. 0 will make the card autosense the ringspeed and join at the appropriate speed, this will be the default option for most people. 4 or 16 allow you to explicitly force the card to operate at a certain speed. The card will fail if you try to insert it at the wrong speed. (Although some hubs will allow this so be *very* careful). The main purpose for explicitly setting the ring speed is for when the card is first on the ring. In autosense mode, if the card cannot detect any active monitors on the ring it will open at the same speed as its last opening. This can be harardous if this speed does not match the speed you want the ring to operate at.

  • pkt_buf_sz: This is this initial receive buffer allocation size. This will default to 4096 if no value is entered. You may increase performance of the driver by setting this to a value larger than the network packet size, although the driver now re-sizes buffers based on MTU settings as well.

  • message_level: Controls level of messages created by the driver. Defaults to 0 which only displays start-up and critical messages. Presently any non-zero value will display all soft messages as well. NB This does not turn debugging messages on, that must be done by modified the source code.

Multi-card. The driver will detect multiple cards and will work with shared interrupts, each card is assigned the next token ring device, i.e. tr0 , tr1, tr2. The driver should also happily reside in the system with other drivers. It has been tested with ibmtr.c running. I have had multiple cards in the same system, all sharing the same interrupt and working perfectly fine together.

Variable MTU size:. The driver can handle a MTU size upto either 4500 or 18000 depending upon ring speed. The driver also changes the size of the receive buffers as part of the mtu re-sizing, so if you set mtu = 18000, you will need to be able to allocate 16 * (sk_buff with 18000 buffer size) call it 18500 bytes per ring position = 296,000 bytes of memory space, plus of course anything necessary for the tx sk_buff's. Remember this is per card, so if you are building routers, gateway's etc, you could start to use a lot of memory real fast.

3.1.6. SysKonnect adapters

Information for the SysKonnect Token Ring ISA/PCI Adapter is courtesy Jay Schulist

The Linux SysKonnect Token Ring driver works with the SysKonnect TR4/16(+) ISA, SysKonnect TR4/16(+) PCI, SysKonnect TR4/16 PCI, and older revisions of the SK NET TR4/16 ISA card.

Latest information on this driver can be obtained on the Linux-SNA WWW site. Please point your browser to: http://www.linux-sna.org

Important information to be noted:

  • 1. Adapters can be slow to open (~20 secs) and close (~5 secs), please be patient.

  • 2. This driver works very well when autoprobing for adapters. Why even think about those nasty io/int/dma settings of modprobe when the driver will do it all for you!

This driver is rather simple to use. Select Y to Token Ring adapter support in the kernel configuration. A choice for SysKonnect Token Ring adapters will appear. This drives supports all SysKonnect ISA and PCI adapters. Choose this option. I personally recommend compiling the driver as a module (M), but if you you would like to compile it staticly answer Y instead.

This driver supports multiple adapters without the need to load multiple copies of the driver. You should be able to load up to 7 adapters without any kernel modifications, if you are in need of more please contact the maintainer of this driver.

Load the driver either by lilo/loadlin or as a module. When a module using the following command will suffice for most:
		# modprobe sktr
		
This will produce output similar to the following: (Output is user specific)
		sktr.c: v1.01 08/29/97 by Christoph Goos
		tr0: SK NET TR 4/16 PCI found at 0x6100, using IRQ 17.
		tr1: SK NET TR 4/16 PCI found at 0x6200, using IRQ 16.
		tr2: SK NET TR 4/16 ISA found at 0xa20, using IRQ 10 and DMA 5.
		
Now just setup the device via ifconfig and set and routes you may have. After this you are ready to start sending some tokens.

Errata. For anyone wondering where to pick up the SysKonnect adapters please browse to http://www.syskonnect.com

Below is the setting for the SK NET TR 4/16 ISA adapters
                    ***************************
                    ***   C O N T E N T S   ***
                    ***************************

                1) Location of DIP-Switch W1
                2) Default settings
                3) DIP-Switch W1 description


		==============================================================
		CHAPTER 1     LOCATION OF DIP-SWITCH
		==============================================================

		+------------------------------------------------------------------+
		|+------+                         +-----+            +---+         |
		||------|                      W1 +-----+     +----+ |   |         |
		||------|                                     |    | |   |      +---+
		||------|              +-----------+          +----+ |   |      |  ||
		||------|              |           |          +---+  +---+      +---+
		||------|              | TMS380C26 |          |   |                |
		||------|              |           |          +---+                |-+
		|+------+              |           |                               | |
		|                      +-----------+                               | |
		|                                                                  | |
		|                                                                  |-+
		|                                                                  |
		|                                                                  |
		|                                                                  |
		|                                                                  |
		+------------+----------------+--+-----------------------+---------+
		             +----------------+  +-----------------------+
		
	
		==============================================================
		CHAPTER 2     DEFAULT SETTINGS
		==============================================================

          W1    1  2  3  4  5  6  7  8
        +------------------------------+
        | ON    X                      |
        | OFF      X  X  X  X  X  X  X |
        +------------------------------+

        W1.1 = ON               Adapter drives address lines SA17..19
        W1.2 - 1.5 = OFF        BootROM disabled
        W1.6 - 1.8 = OFF        I/O address 0A20h
		
		==============================================================
		CHAPTER 3     DIP SWITCH W1 DESCRIPTION
		==============================================================

      +---+---+---+---+---+---+---+---+  ON
      | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
      +---+---+---+---+---+---+---+---+  OFF
      |AD | BootROM Addr. |  I/O      |
      +-+-+-------+-------+-----+-----+
        |         |             |
        |         |             +------ 6     7     8
        |         |                     ON    ON    ON       1900h
        |         |                     ON    ON    OFF      0900h
        |         |                     ON    OFF   ON       1980h
        |         |                     ON    OFF   OFF      0980h
        |         |                     OFF   ON    ON       1b20h
        |         |                     OFF   ON    OFF      0b20h
        |         |                     OFF   OFF   ON       1a20h
        |         |                     OFF   OFF   OFF      0a20h    (+)
        |         |
        |         |
        |         +-------- 2     3     4     5
        |                   OFF   x     x     x       disabled  (+)
        |                   ON    ON    ON    ON      C0000
        |                   ON    ON    ON    OFF     C4000
        |                   ON    ON    OFF   ON      C8000
        |                   ON    ON    OFF   OFF     CC000
        |                   ON    OFF   ON    ON      D0000
        |                   ON    OFF   ON    OFF     D4000
        |                   ON    OFF   OFF   ON      D8000
        |                   ON    OFF   OFF   OFF     DC000
        |
        |
        +----- 1
               OFF    adapter does NOT drive SA<17..19>
               ON     adapter drives SA<17..19>  (+)


        (+) means default setting

		

3.1.7. PCMCIA

3.1.7.1. Introduction

PCMCIA Token Ring adapters will work on all versions of the Linux kernel. Unfortunately, the road to hell is often paved with melting snowballs ;-) and there are a myriad of different combinations that can be used to get the adapters to work, all with different options, different requirements and different issues. Hopefully with this document you will be able to figure out which combinations of ingredients are required and how to get them up and running on your machine.

3.1.7.2. History

In the 2.0.x and 2.2.x kernels days, pcmcia was only available as an external package, created and maintained by David Hinds. When the only stable kernel available was 2.0.36, life was pretty easy and with a few simple configuration options the adapters would work.

With the advent of 2.2.x, ibmtr.c was completely updated, which broke the pcmcia driver (ibmtr_cs.c). The pcmcia driver was updated to work with the new ibmtr driver and the 2.2.x kernels. This is where the first level of complication starts. As the pcmcia_cs package is stand alone, it has to support the various different kernels, so instead of being able to have different versions of drivers in different versions of the kernel source, the pcmcia_cs drivers must work with all kernel versions. This not only creates some ugliness in the driver itself but also causes confusion as to which version of pcmcia_cs works for the latest kernel.

At this point, everything was working fine, and then come along the 2.3.x develpment series of kernels. The 2.3.x kernels provided their own support for pcmcia and the ibmtr_cs driver was included in the kernel proper. So now there were two ways of getting pcmcia token ring support, either using the kernel drivers themselves or using the pcmcia_cs package, not too much of a problem because only developers were using the 2.3.x kernels. Of course this all changed when the 2.4 kernel was released and a lot more users started using the kernel.

During late 2000, early 2001, significant development work was done on both the standard ibmtr driver and the pcmcia driver. Original pcmcia updates including using high memory and hot-eject support. These initial updates were only for the 2.2.x kernels, and hence only included in the pcmcia_cs package. Later development saw great improvements in ibmtr and ibmtr_cs for the 2.4.x kernels. So as of writing, 1/23/02 , there are many different combinations of kernel version and driver floating around especially considering that different distributions have released different versions of the 2.4 kernels.

3.1.7.3. 2.0.x kernels

If you are using one of the 2.0.x kernels, then I salute your perserverance and really you should have got the pcmcia drivers configured and working by now ;-)

You will have to use the pcmcia_cs package and play with the /etc/pcmcia/config.opts, see the section below about config.opts fun. Just about any version of pcmcia_cs that's been released in the last 2/3 years will work fine.

3.1.7.4. 2.2.0 - 2.2.6 kernels

These were the series of kernels where the pcmcia driver didn't work at all. It's probably just easiest to upgrade the kernel to a later version.

If you really do need to get this up and running, then a recent pcmcia_cs is required and you should be able to grab the ibmtr.c and ibmtr.h from a 2.2.7 - 2.2.16 kernel and use them (note no greater than 2.2.16 !!)

You have to do the config.opts mangling, see the section on setting all this up.

3.1.7.5. 2.2.7 - 2.2.16 kernels

These kernels are well supported, simply use the pcmcia_cs package and play with the config.opts file.

3.1.7.6. 2.2.17 - 2.2.19 kernels

The pcmcia driver was updated for these kernel to eliminate the need for the config.opts mangling. You'll need pcmcia_cs at least 3.1.24, although it is probably better just to grab the latest version.

Simply compile up pcmcia_cs and you're done. No need to play with config.opts, in fact if you've been running a previous version that did have the ibmtr_cs line in config.opts it would be a very good idea to remove or comment out the line. The new driver allocates the entire 64k for shared ram and it needs to be aligned on a 64k boundary, if you've got a previous srambase value not on a 64k boundary, the driver will barf and the kernel will panic.

3.1.7.7. 2.4.0 - 2.4.4 (non Redhat) kernels

Use the built-in kernel pcmcia driver and play with config.opts.

If you want to use the latest and greatest version of the driver with the high memory and hot-swap support you can download the patch and patch up your kernel. Then the line in config.opts can be removed and everything will work fine.

3.1.7.8. 2.4.4-ac11 > kernels

These kernels include the new drivers so simply compile up the drivers, ensure that there is no configuration line in config.opts and away you go.

3.1.7.9. 2.4.2 mangled, i.e. Redhat 7.1

When RedHat released 7.1 with the 2.4.2 kernel they modified the kernel (as they always do) and included the updated ibmtr/ibmtr_cs driver from the web site. If you're lucky this may work straight out of the box (again no need for the ibmtr_cs line in config.opts), if not then it is probably easiest to upgrade to the latest 2.4.x kernels and use the drivers there. (The reason being that while I will work out how to get around a distribution caused problem, I will not provide support for them, I'll answer questions and give help because I'm a nice guy, but I am not going to provide driver updates against distributions. Official support is for the drivers in the kernels available from the official kernel mirrors.

3.1.7.10. 2.4.x kernels and pcmcia_cs

There is no need to use pcmcia_cs with the 2.4 kernels to get the token ring adapters up and running, but I appreciate that some of you may need to use pcmcia_cs to get other adapters working that are not supported properly in the kernel.

The pcmcia_cs package will not work with the latest drivers, it may work with the 2.4.0-2.4.4 drivers. I am currently in two minds about providing support with pcmcia_cs for the 2.4 kernels, you can ask me directly or check the web site every now and then so see if anything has changed.

3.1.7.11. Config.opts mangling (or how to send yourself insane)

This is the hardest part to getting the pcmcia adapters working with the drivers that need the ibmtr_cs line in /etc/pcmcia/config.opts. No set of values is guaranteed to work the same on a different machine. It really is a case of trial and error but forewarned and forearmed with a little bit of knowledge can make the process a whole lot easier.

"Hey, I don't care, just give me something that works"

OK, try this, it works in most situations, if it doesn't you have to read the rest of the section anyway. Just insert the following line in /etc/pcmcia/config.opts
modules "ibmtr_cs" opts "mmiobase=0xd2000 srambase=0xd4000"
restart pcmcia and insert the adapter.

"OK, that didn't work, bring on the pain"

The pcmcia driver need to allocate two areas of memory to operate properly. All areas of memory allocated must be aligned on the same boundary as the size of the area being aligned, i.e. a block 8K in size must be on an 8K boundary (0xc8000, 0xca000, 0xcc000, 0xce000, 0xd0000, 0xd2000) and for a 16K block must be on a 16K boundary (0xc8000, 0xcc000, 0xd0000, 0xd4000). All memory areas must be allocated within the ISA address space, 0xC0000-0xDFFFF). Theoretically you should be able to use anywhere within this area, although experience has shown that most machines hide stuff in the 0xc0000-0xc9fff area. Some machines have even been known to use the 0xd0000-0xd1fff area without telling anybody (some thinkpads !!). So you really want to stick with memory allocations in the 0xcc000 - 0xdffff range.

Of course, the two memory areas cannot overlap either ;)

The first area of memory is an 8K area for the memory mapped input/output (MMIO) and must be placed on an 8K boundary. This area of memory is not usually the cause of any problems and can be placed pretty much anywhere, recommended values are: 0xcc000, 0xd0000,0xd2000,0xd4000.

The second area of memory can be sized to fit your desires, this is the area of memory where the incoming and outgoing packets are stored and received. The driver defaults to a 16K memory size and must be placed on a 16K boundary. Good areas are: 0xd0000,0xd4000,0xd8000.

Once you've decided which areas of memory you are goin to try, you need to add the correct line to the /etc/pcmcia/config.opts file. Configuration lines in this file take the format of:
			module "module_name" opts "option1=opt1_value option2=opt2_value ...."
			
In our case module_name is ibmtr_cs. There are three options that be set with the ibmtr_cs driver, mmiobase, srambase and sramsize.

If they are not set they will revert to the defaults in the driver, which in 9 cases out of 10 won't work for you. sramsize rarely has to be set unless you are looking for that last little bit of performance from your adapter.

So, having decided upon your values, let's say 0xd2000 for the MMIO and 0xd4000 for the shared memory you would build a config.opts line like this:
			module "ibmtr_cs" opts "mmiobase=0xd2000 srambase=0xd4000" 
			
The pcmcia_cs package must be restarted for these new options to take effect, usually with:
/etc/init.d/pcmcia restart or /etc/rc.d/init.d/pcmcia/restart
depending upon which run level organization your distribution adheres to.

Then just plug it in and see if it works. If not you'll just have to go back and change the values for mmiobase and srambase until you find a combination that works. Or, you can upgrade to a kernel/pcmcia_cs version that support high memory allocation, where all this config.opts nonsense is not required and you can just happily plug your adapter in and watch it run.

3.1.8. Madge Supplied Drivers

Madge released 2.31 of their driver in 1999 and 2.41 in late 2001. Both drivers can be downloaded from the Madge web site and the 2.41 driver is also available from the Linux Token Ring Project web site.

Once the drivers have been downloaded, see the README file that comes with the drivers for instruction on how to built and install the drivers. The only other issue some people find with the drivers is a failure to build the tool chain due to an incorrect version of the newt libraries. If you get a compiler error relating to newt.h change the madge-source/include/mtok/config.h file so that the #define NEWNEWT line reads:
		#define NEWNEWT 1
		
This will ensure the tools use the correct newt libraries during the build process.

A patch is available from the Linux Token Ring Project web site for the 2.31 drivers to enable them to work with the 2.4.x kernels.

3.1.9. Olicom Drivers

Back when Olicom were still in business they did produce a Linux driver that does actually work. Trying to find the driver these days is a bit tough. If the ftp.olicom.com site is still up and running, the driver can be found there.

The driver is a combination of GPL source code and proprietary binary low level code. The driver only works with the 2.0.36 and 2.2.x kernels. It should be possible to port this driver to the 2.4.x kernels...