DP Tool Club 15/CD_ASCQ_15_070894.iso/vrac/how_to.zip/ETHER1.HOW

Text File | 1994-04-29 | 63KB | 1,507 lines

Archive-Name: linux/howto/ethernet/part1
Last-modified: 15 Apr 94

Linux Ethernet HOWTO v1.0 -- Last updated Mar 16, 1994
=================================================================

-- covers changes up to and including Linux kernel v1.0

INDEX:

0 Introduction.
0.01  How do I use this Guide?
0.01  Disclaimer and Copyright
0.02  Questions already?
0.03  Related Documentation
0.04  New Versions of this Document
0.05  Feedback

1 What card should I buy for Linux?
1.01  Eight bit vs 16 bit
1.02  Low price Ethernet cards
1.03  Vendors and brands to avoid.
1.04  Type of cable that your card should support

2 Status of various Ethernet cards under Linux.
2.01  3Com
2.02  Western Digital / SMC
2.03  NExxxx
2.04  Hewlett Packard Cards
2.05  D-Link
2.06  Cabletron
2.07  Allied Telesis
2.08  Arcnet
2.09  Digital / DEC
2.10  Intel
2.11  PureData
2.12  Xircom
2.13  Zenith
2.14  Racal-Interlan
2.15  AMD LANCE (79C960)
2.16  AT-Lan-Tec / RealTek Pocket adaptor
2.17  Ansel
2.18  DFI

3 Clones of popular Ethernet cards.
3.01  WD80x3 Clones
3.02  NE2000 Clones

4 Cables, coax, twisted pairs etc.
4.01  Thin Ethernet (thinnet)
4.02  Twisted Pair
4.03  Thick Ethernet

5 Technical information.
5.01  Probed addresses
5.02  Skeleton / prototype driver
5.03  Driver interface to the kernel
5.04  Interrupts and linux
5.05  Programmed I/O vs. shared mem. vs slave/master DMA
5.06  Programming the Intel chips (i82586 and i82593)
5.07  Programming information from 3Com
5.08  Notes on AMD PCnet-ISA / LANCE Based cards (79C960)
5.09  Multicast and Promiscuous mode
5.10  The Berkeley Packet Filter (BPF)
5.11  Unresolved questions / concerns

6 Possible problems, questions and troubleshooting.
6.01  Problems with NE2000 (and clones)
6.02  Problems with WD80*3 cards
6.03  Problems with 3Com cards

7 Networking with a laptop computer.
7.01  Option 1 -- using SLIP
7.02  Option 2 -- Built in NE2000 compatible or PCMCIA Ethercard.
7.03  Option 3 -- ISA Ethercard in the docking station.
7.04  Option 4 -- Pocket / parallel port adaptors.

8 Frequently asked questions.
8.01  Just the FAQ's ma'am -- just the FAQ's.

9 Miscellaneous.
9.01  Bad Vendors
9.02  Closing

======================================================================

0. Introduction.

 This is the Ethernet-HOWTO, which is a compilation of information
 about which ethernet devices can be used for Linux, and how to
 set them up.

 This Ethernet-HOWTO is by:
  Donald J. Becker <[e-mail]>
  Paul Gortmaker  <[e-mail]>

 It covers what cards you should and shouldn't buy; how to set
 them up, how to run more than one, and other common problems and
 questions. It does *not* cover the software end of things, as that
 is covered in the NET-2 HOWTO.

 Other people who have contributed (directly or indirectly) are,
 in alphabetical order:

 Peter Bauer  <[e-mail]>
 Ross Biro  <[e-mail]>
 Alan Cox  <[e-mail]>
 David C. Davies  <[e-mail]>
 Bjorn Ekwall  <[e-mail]>
 Charles Hedrick  <[e-mail]>
 Mike Jagdis  <[e-mail]>
 Duke Kamstra  <[e-mail]>
 Russell Nelson  <[e-mail]>
 Cameron Spitzer  <[e-mail]>
 Dave Roberts  <[e-mail]>
 Glenn Talbott  <[e-mail]>
 Miquel van Smoorenburg <[e-mail]>

 Many thanks to the above people, and all the other unmentioned
 testers out there.

0.01 How Do I Use This Guide?

 As this guide is getting bigger and bigger, you probably don't want
 to spend the rest of your afternoon reading the whole thing. And you
 don't *have* to read it all. If you haven't got an ethernet card, then
 you will want to start with section one to see what you should buy,
 and what you should avoid. If you have already got an ethernet card,
 but are not sure if you can use it with Linux, then you will want to
 read section two, which contains specific information on each
 manufacturer, and their cards. If you are having trouble with your
 card, then you will want to read the specific information about
 your card in section two and the troubleshooting information in
 section six. If you are interested in some of the technical aspects
 of the device drivers, then you can find that information in
 section 5.

0.01 Disclaimer and Copyright

 This document is *not* gospel. However, it is probably the most
 up to date info that you will be able to find. Nobody is responsible
 for what happens to your hardware but yourself. If your ethercard
 or any other hardware goes up in smoke (...nearly impossible!)
 we take no responsibility. ie. THE AUTHORS ARE NOT RESPONSIBLE
 FOR ANY DAMAGES INCURRED DUE TO ACTIONS TAKEN BASED ON THE
 INFORMATION INCLUDED IN THIS DOCUMENT.

 This document is Copyright (c) 1994 by Donald Becker and
 Paul Gortmaker. Permission is granted to make and distribute
 verbatim copies of this manual provided the copyright notice
 and this permission notice are preserved on all copies.

 Permission is granted to copy and distribute modified versions
 of this document under the conditions for verbatim copying,
 provided that this copyright notice is included exactly as in
 the original, and that the entire resulting derived work is
 distributed under the terms of a permission notice identical
 to this one.

 Permission is granted to copy and distribute translations
 of this document into another language, under the above
 conditions for modified versions.

0.02 Questions already?

 If you have questions about your ethernet card, please READ this
 document first. You may also want to join the NET channel of the
 Linux-activists mailing list by sending mail to
  [e-mail]
 with the line
  X-Mn-Admin: join NET
 at the top of the message body (not the subject). If you want to
 learn how to use the mailing channels, then send an empty message
 to the above address, and you will get an instruction manual sent
 back to you in a few hours. However, it is worth noting that the NET
 channel is primarily used for discussion of the networking code, and
 you may not see much discussion about a particular driver.
 Furthermore keep in mind that the NET channel is for development
 discussions only. General questions on how to configure your system
 should be directed to comp.os.linux.help unless you are actively
 involved in the development of part of the networking for Linux.
 We ask that you *please* respect this general guideline for content.
 You can safely bet that neither of the authors will respond to
 any plea for help that *should* be posted to c.o.l.help, but is
 inappropriately placed elsewhere.

0.03 Related Documentation

 Much of this info came from saved postings from the comp.os.linux
 groups, which shows that it is a valuable resource of information.
 Other useful information came from a bunch of small files by Donald
 himself. Some of these are found at /pub/linux/info on ftp.super.org
 [192.31.192.1] Of course, if you are setting up an Ethernet card,
 then you will want to read the NET-2 HOWTO so that you can actually
 do something with it. ftp.super.org is the home of most alpha drivers
 that are not presently in the kernel. And last but not least, the
 contributions from the individuals and companies listed above are
 greatly appreciated as well. Oh yeah, if you fancy yourself as
 a bit of a hacker, you can always scrounge some additional info
 from the driver source files as well. There is usually a paragraph
 in there describing any important points.

0.04 New versions of this document

 New versions of this document can be retrieved via anonymous
 FTP from sunsite.unc.edu:/pub/Linux/docs/HOWTO/* and various
 Linux ftp mirror sites. It will also be posted to the newsgroup
 comp.os.linux.announce at a regular interval. Updates will be made
 as new information / drivers becomes available. If this copy
 that you are reading is more than 2 months old, it is either out of
 date, or it means that I have been lazy and haven't updated it.

0.05 Feedback

 Any corrections can be sent to one of us ([e-mail]
 or [e-mail]) We will *attempt* to keep this up to date as
 more drivers become available, and as the networking code matures.

1 What card should I buy for Linux?

 For impatient users that just want a quick, cheap answer the
 summary is: get 16 bit thinnet 8013 cards. For more detail as
 to the who what where and why, read on.

1.01 Eight bit vs 16 bit

 Unless you are a light user, or are confined to using the smaller
 ISA slot, the use of the 8 bit cards like the wd8003 and the 3c503
 is really not worth the cost savings. Get the 8013 or the 3c503/16
 instead.

1.02 Low price Ethernet cards

 I keep track of the current low-price vendors, just because it's
 asked so often. Call AT-LAN-TEC at 301-948-7070. Ask for their
 technical support person, "Vincent Bono". As with all purchases,
 you should indicate you are buying this for a Linux system.
 The last I checked the price for 10 NE2000s was $480, or $48 ea.!
 NB: Their current NE2000 clone is a model that "traps" other
 drivers that probe into their address space. AT-LAN-TEC also carries
 a clone, non-EEPROM 8013 board for somewhat more, and a NE2100 clone.
 Either is a better choice if the very lowest price isn't essential.
 Also, SMC is offering an evaluation deal on their new Ultra cards,
 and the word is that you can get one for $50. You can ask them
 yourself by calling 1-800-SMC-4YOU in Canada and the USA.

 The Allied Telesis AT1500 is offered at a good price by many vendors.
 Even Inmac, known for their premium markup, has this card for under
 $100.

1.03 Vendors and Brands to Avoid

 These vendors have decided *not* to release programming information
 about their products, without signing a non-disclosure agreement.
 More information can be found in section two and 9.01. Hence it is
 strongly advised that you avoid buying products offered from
 these companies.

  (1) Cabletron
  (2) Xircom

 These particular cards should be avoided, as they are obsolete.
 The reasons as to why they have been classified as such can be
 found in section 2 of this document.

  (1) 3c501
  (2) Arcnet

1.04 Type of cable that your card should support

 Unless you have to conform to an existing network, you will want
 to use thinnet or thin ethernet cable. This is the style with the
 standard BNC connectors. See section 4 for other concerns with
 different types of ethernet cable.

 Most ethercards also come in a "Combo" version for only $10-$20 more.
 These have both twisted pair and thinnet transceiver built-in,
 allowing you to change your mind later.

2 Status of Various Ethernet Cards under Linux

 The only thing that one needs to use an ethernet card with Linux
 is the appropriate driver. For this, it is essential that the
 manufacturer will release the technical programming information to
 the general public without you (or anyone) having to sign your life
 away. A good guide for the likelihood of getting documentation
 (or, if you aren't writing code, the likelihood that someone
 else will write that driver you really, really need) is the
 availability of the Crynwr (nee Clarkson) packet driver. Russ
 Nelson (see the acknowledgements in the intro.) runs this
 operation, and has been very helpful in supporting the development
 of drivers for Linux.

 Given the documentation, you can write a driver for
 your card and use it for Linux, at least in theory. Keep in
 mind that some old hardware that was designed for XT type
 machines will not function very well in a multitasking
 environment such as Linux. Use of these will lead to major
 problems if your network sees a reasonable amount of traffic.

 Most cards come with drivers for MS-DOS interfaces such as
 NDIS and ODI, but these are useless for Linux. Many people
 have suggested directly linking them in or automatic
 translation, but this is nearly impossible. The MS-DOS
 drivers expect to be in 16 bit mode and hook into "software
 interrupts", both incompatible with the Linux kernel. This
 incompatibility is actually a feature, as some Linux drivers
 are considerably better than their MS-DOS counterparts. The
 "8390" series drivers, for instance, use ping-pong transmit
 buffers, which are only now being introduced in the MS-DOS world.

 Keep in mind that PC ethercards have the widest variety of
 interfaces (shared memory, programmed I/O, bus-master, or slave
 DMA) of any computer hardware for anything, and supporting a
 new ethercard sometimes requires re-thinking most of the lower-level
 networking code. (If you are interested in learning more about
 these different forms of interfaces, see section 5)

 Also, similar product numbers don't always indicate similar products.
 For instance, the 3c50* product line from 3Com varies wildly
 between different members.

 Enough talk. Let's get down to the information you want.

2.01 3Com

 Supported:

  3c503, 3c503/16
   3Com shared-memory ethercards. They also have a
   programmed I/O mode that doesn't use the 8390
   facilities (their engineers found too many bugs!)
   It should be about the same speed as the same bus
   width WD80x3, but I don't have a 16 bit version
   to benchmark. Unless you are a light user, spend
   the extra money and get the 16 bit model, as the
   price difference isn't significant. The 3c503 does not
   have "EEPROM setup", so the diagnostic/setup program
   isn't needed before running the card with Linux. The
   shared memory address of the 3c503 is set using jumpers
   that are shared with the boot PROM address. This is
   confusing to people familiar with other ISA cards,
   where you always leave the jumper set to "disable"
   unless you have a boot PROM.

   The Linux 3c503 driver can also work with the 3c503
   programmed-I/O mode, but this is slower and less
   reliable than shared memory mode. Also, programmed-I/O
   mode is not tested when updating the drivers, the
   deadman (deadcard?) check code may falsely timeout on
   some machines, and the probe for a 3c503 in
   programmed-I/O mode is turned off by default in some
   versions of the kernel. This was a panic reaction to
   the general device driver probe explosion; the 3c503
   shared memory probe is a safe read from memory, rather
   than an extensive scan through I/O space. As of pl13,
   the kernel has an I/O port registrar that makes I/O
   space probes safer, (see section 5.1 for more info.)
   and the programmed-I/O 3c503 probe has been re-enabled.
   You still shouldn't use the programmed-I/O mode though,
   unless you need it for MS-DOS compatibility.

   The 3c503's IRQ line is set in software, with no hints
   from an EEPROM. Unlike the MS-DOS drivers, the
   Linux driver has capability to autoIRQ: it uses the
   first available IRQ line in {5,2/9,3,4}, selected each
   time the card is 'ifconfig'ed. (Older driver versions
   selected the IRQ at boot time.) The ioctl() call
   in 'ifconfig' will return EAGAIN if no IRQ line is
   available at that time.

  3c509
   A fairly new card from 3Com. It's inexpensive and has
   excellent performance for a non-bus-master design. The
   drawbacks are that it _requires_ very low interrupt
   latency, and it isn't rated for bus speeds greater than
   8Mhz.

   A working 3c509 driver was first included as an
   alpha-test version in the 0.99pl13 kernel sources.
   It is now in the standard kernel.

   The 3c509 has a tiny Rx buffer, causing the driver to
   occasionally drop a packet if interrupts are masked for
   too long. To minimize this problem, the driver should
   be completely rewritten to use predictive interrupts.
   (Note: performance re-writes of working drivers are low
   priority unless there is some particular incentive or
   need.)

   There is also an alpha version of a Linux 3c509
   diagnostic and EEPROM setup program, but for now
   users that don't like the defaults should use the
   MS-DOS EEPROM setup program.

  3c579
   The EISA version of the 509. The current EISA version
   uses the same 16 bit wide chip rather than a 32 bit
   interface, so the performance increase isn't stunning.
   The EISA probe code was added to 3c509.c for pl14.
   We would be interested in hearing progress reports
   from any 3c579 users. (Read the above 3c509
   section for info on the driver.)

   Cameron Spitzer writes:
   "The 3C579 (Etherlink III EISA) should be configured
   as an EISA card. The IO Base Address (window 0
   register 6 bits 4:0) should be 1f, which selects EISA
   addressing mode. Logic outside the ASIC decodes the
   IO address s000, where s is the slot number. I don't
   think it was documented real well. Except for its IO
   Base Address, the '579 should behave EXACTLY like the
   '509 (EL3 ISA), and if it doesn't, I want to hear
   about it (at my work address).

   I will leave it to the Real Programmers to suggest
   the right hack to /usr/src/linux/net/inet/3c509.c to
   take care of the EISA case.

   (Note that the drivers now reside in ./drivers/net/
    and *not* ./inet/net/ --- pg.)

   Beware that if you put a '509 in EISA addressing mode
   by mistake and save that in the EEPROM, you'll have
   to use an EISA machine or the infamous Test Via to
   get it back to normal, and it will conflict at IO
   location 0 which may hang your ISA machine. It's not
   my job to say whether this is a bug or feature, but I
   have heard loud and clear that customers don't like
   it and I don't think we'll do it that way again."

 Unsupported:

  3c501
   Too brain-damaged to use. Available surplus from many
   places. Avoid it like the plague. Again, do not
   purchase this card, even as a joke. It's performance
   is horrible, and it breaks in many ways.

   (I have a standing offer: I'll pay $2 for each 3c501
   shipped to me postpaid, but only if you include the
   BNC 'T' connector and the jumpers. $2.50 if you just
   send the 'T', jumpers, and address PROM and promise to
   destroy the board. -djb)

   Cameron L. Spitzer of 3Com said:
   "I'm speaking only for myself here, of course, but I
   believe 3Com advises against installing a 3C501 in a
   new system, mostly for the same reasons Donald has
   discussed. You probably won't be happy with the
   3C501 in your Linux box. The data sheet is marked
   "(obsolete)" on 3Com's Developers' Order Form, and
   the board is not part of 3Com's program for sending
   free Technical Reference Manuals to people who need
   them. The decade-old things are nearly
   indestructible, but that's about all they've got
   going for them any more."

   For those not yet convinced, the 3c501 can only do one
   thing at a time -- while you are removing one packet
   from the single-packet buffer it cannot receive
   another packet, nor can it receive a packet while are
   loading a transmit packet. This was fine for a
   network between two 8088-based computers where
   processing each packet and replying took 10's of
   msecs, but modern networks send back-to-back
   packets for almost every transaction.

   Having read this far, you must be persistent, so you
   get let in on a secret. As of pl13, some more of
   the hardware problems were "compensated for".

   Ie. in a fit of madness I wasted a whole day updating
   my 3c501 driver and then trying to track down a few
   more of the 3c501 glitches. It now works well enough
   to NFS mount filesystems, but the receiver still
   occasionally hangs. I'm mostly certain that this is
   a hardware bug. When it hangs, the next set of
   outgoing packets will reset the board, but that's
   only useful if you have something occasionally
   generating outgoing packets.

   The driver is now in the std. kernel, but under the
   following conditions: This is unsupported code. I
   know my usual copyright says all the code is
   unsupported, but this is _really_ unsupported. I
   DON'T want to see bug reports, and I'll accept bug
   fixes only if I'm in a good mood that day.

   I don't want to see a fest of "Linux ethercards for
   sale" postings. A bunch of people have bought dozens
   of "dumpster special" 3c501s, and they hope to sell
   them at rip-off prices. A 3c501 is barely worth the
   shipping cost, and if I see people trying to sell
   them here by claiming "supported by Linux" I _will_
   flame them. They are _not_ supported by Linux.

   I don't want to be flamed later for putting out bad
   software. I don't know all all of the 3c501 bugs,
   and I know this driver only handles a few that I've
   been able to figure out. It has taken a long
   intense effort just to get the driver working this
   well.

   That said, you will find it included in "config.in"
   No special mods are needed to use it with pl15
   or greater kernels. Jumper your card to 0x280.

   AutoIRQ works, DMA isn't used, the autoprobe only
   looks at 0x280, the debug level is set with the third
   boot-time argument. You'll probably want to change
   the default EL_DEBUG to '2'.

   Once again, THE USE OF A 3c501 IS STRONGLY DISCOURAGED
   and it is NOT SUPPORTED BY LINUX.


  3c505
   An Intel-based ethercard with no driver available
   at present. (Not a very common card.)

  3c507
   This card uses one of the Intel chips, and the
   development of the driver is closely related to
   the development of the Intel Ether Express driver.
   The driver has been included in the standard
   release of pl15. You will have to un-comment
   the 3c507 line in "config.in" -- in case you
   didn't figure it out already, it is commented
   out because it is still being tested.

   Technical information is available in section 5.06,
   and if you have experience in writing drivers, see
   section 5.07 as well.

2.02 Western Digital / SMC

 The ethernet part of Western Digital has been bought by SMC. The
 SMC Elite and SMC Elite Plus are the same as late-model WD8003
 and WD8013 cards. Note that the SMC Elite Ultra is *not* the
 same as the plain SMC Elite / WD8013 card. (see below)

 Supported:

  WD8003, WD8013, SMC Elite, SMC Elite Plus
   A shared memory design by Western Digital. The
   8 bit 8003 is slightly less expensive, but only
   worth the savings for light use. Over the
   years the design has added more registers and an
   EEPROM. Clones usually go by the '8013' name, and
   usually use a non-EEPROM (jumpered) design. This part
   of WD has been sold to SMC, so you'll usually see
   something like SMC/WD8013 or SMC Elite Plus (WD8013).
   The shared memory makes the cards 10-20% faster,
   especially with larger packets. More importantly
   (to me at least) it avoids a few bugs in the
   programmed-I/O mode of the 8390, allows safe
   multi-threaded access to the packet buffer, and
   doesn't have a programmed-I/O data register that
   hangs your machine during warm-boot probes.

  SMC Elite 16 ULTRA
   This ethercard is based on a new chip from SMC, with
   a few new features. While it has a mode that is
   similar to the older SMC ethercards, it's not
   compatible with the old WD80*3 drivers. However, in
   this mode it shares most of its code with the other
   8390 drivers, while operating somewhat faster than a
   WD8013 clone.

   Some of the device probe checks in pl14 were too
   too strict, causing some cards to not be detected
   every time. This was fixed for pl14a, and hence is
   fine for pl15. Since part of the Ultra "looks" like
   an 8013, the Ultra probe is supposed to find an
   Ultra before the wd8013 probe has a chance to
   mistakenly identify it.

   Std. as of pl14, and made possible by documentation
   and ethercard loan from [e-mail],
   Duke Kamstra. If you plan on using an Ultra with Linux
   send him a note of thanks to let him know that there
   are Linux users out there!

   I'm considering writing a separate driver for the
   Ultra's "Altego" mode which allows chaining transmits
   at the cost of inefficient use of receive buffers,
   but that will probably not happen right away.
   Performance re-writes of working drivers are low
   priority unless there is some particular incentive
   or need.

2.03 NExxxx

 The prefix "NE" came from Novell Ethernet. Novell followed the
 cheapest NatSemi databook design and sold the manufacturing rights
 (spun off?) Eagle, just to get reasonably-priced ethercards into
 the market.

 Supported:

  NE1000, NE2000
   The now-generic name for a bare-bones design around
   the NatSemi 8390. They use programmed I/O rather than
   shared memory, leading to easier installation but
   slightly lower performance and a few problems. Again,
   the savings of using an 8 bit NE1000 over the NE2000
   are only warranted if you expect light use. Some
   recently introduced NE2000 clones use the National
   Semiconductor "AT/LANTic" 83905 chip, which offers
   a shared memory mode similar to the 8013 and EEPROM
   or software configuration. Some problems can arise
   with poor clones. See the question and answer section
   later in this document, and the section on clones.
   I have written a NE2000 diagnostic program, but it
   is still presently in alpha test. (ne2k)

  NE1500, NE2100
   The AT1500 driver, recently added to the list of
   supported cards, also supports the NE1500, NE2100 and
   clones. The driver shipped with pl12 kernel doesn't
   detect non-AT1500 cards with autoprobe, but will work
   fine if you specify the base address explicitly and
   jumper for DMA channel 5. Read the Allied Telesis
   section for more information on LANCE based cards.

2.04 Hewlett Packard

 The 272** cards use programmed I/O, similar to the NE*000 boards,
 but the data transfer port can be "turned off" when you aren't
 accessing it, avoiding problems with autoprobing drivers.

 Thanks to Glenn Talbott for cleaning up the confusion in this
 section regarding the version numbers of the HP hardware, and
 adding lots of new info.

 Supported:

  27245A
   8 Bit 8390 based 10BaseT, not recommended for all the
   8 bit reasons. It was re-designed a couple years
   ago to be highly integrated which caused some
   changes in initialization timing which only
   affected testing programs, not LAN drivers. (The
   new card is not 'ready' as soon after switching
   into and out of loopback mode.)

  27247B, 27252A
   The 47B is a 16 Bit 8390 based 10BaseT w/AUI, and
   the 52A is a 16 Bit 8390 based ThinLAN w/AUI.
   These cards are high performers (3c509 speed) without
   the interrupt latency problems (32K onboard RAM for TX
   or RX packet buffering). They both offer LAN
   connector autosense, data I/O in I/O space (simpler) or
   memory mapped (faster), and soft configuration. 27247B
   was rated Best for ISA Servers by PC Mag this year.

  27247A
   This is the older model that existed before the "B".
   Two versions 27247-60001 or 27247-60002 have part
   numbers marked on the card. Functionally the same to
   the LAN driver, except bits in ROM to identify
   boards differ. -60002 has a jumper to allow
   operation in non-standard ISA busses (chipsets
   that expect IOCHRDY early.)

  HP J2405A
   These are lower priced, and slightly faster than the
   27247B/27252A, but are missing some features, such
   as AUI, ThinLAN connectivity, and boot PROM socket.
   This is a fairly generic LANCE design, but a minor
   design decision makes it incompatible with a generic
   "NE2100" driver. Special support for it (including
   reading the DMA channel from the board) is in pl14
   and up, thanks to information provided by HP's Glenn
   Talbott, [e-mail]. Note that the pre pl14
   driver should not be used with this card.

   More information on LANCE based cards can be found in
   section 5.08.

2.05 D-Link

 Supported:

  DE-600
   Laptop users and other folk who might want a quick
   way to put their computer onto the ethernet may want
   to use this. The driver was included with the default
   kernel source tree as of pl12 and possibly earlier.
   Bjorn Ekwall <[e-mail]> wrote the original.
   Expect about 80kb/s transfer speed from this via the
   parallel port. You should read the README.DLINK
   file in the kernel source tree. The latest release
   of this driver is v0.32, and it is included in the
   standard kernel of pl15

  DE-650
   Some people have been using this PCMCIA card for
   some time now with their notebooks. Note however,
   that using a PCMCIA card with Linux is not trivial.
   See the section on networking with a notebook for
   more information on PCMCIA cards. This driver is
   *not* part of the standard kernel.

  DE-100, DE-200, DE-220-T
   The manual says that it is 100% compatible with the
   NE2000. This is not true. You should call them and
   tell them you are using their card with Linux, and they
   should correct their documentation. Some pre-0.99pl12
   driver versions may have trouble recognizing the DE2**
   series as 16 bit cards, and these cards are the most
   widely reported as having the spurious transfer address
   mismatch errors. Note that there are cards from
   Digital (DEC) that are also named DE100 and DE200,
   but the similarity stops there.

 Unsupported:

  DE-620
   Same as the DE-600, only with two output formats.
   (BNC and RJ-45, I would assume... ????)
   Bjorn writes: "I have still no information on the
   DE-620 that I can include in this release. (Maybe
   someone well connected to D-Link sees this,
   hint, hint, hint...)

2.06 Cabletron

 Yes, another one of these companies that won't release its
 programming information. They waited for months before actually
 confirming that all their information was proprietary, deliberately
 wasting my time. Avoid their cards like the plague if you can.
 Also note that some people have phoned Cabletron, and have been
 told things like "a [e-mail] is working on a driver
 for linux" -- making it sound like I work for them. This is
 NOT the case. Anyway, if I were working for them, or even if
 I had signed a ND agreement, I wouldn't be able to tell
 everyone what a sleazy design the E2100 is. (See below.)

 If you feel like asking them why they don't want to release their
 info so that people can use their cards, write to [e-mail]
 Tell them that you are using Linux, and are disappointed that they
 don't support open systems. (See section 9.01)

 Supported: (...well, not *really* supported)

  E10**, E10**-x, E20**, E20**-x
   These are NEx000 almost-clones that are reported to
   work with the standard NEx000 drivers, thanks to a
   ctron-specific check during the probe. If there are
   any problems, they are unlikely to be fixed, as the
   programming information is unavailable.


  E21**
   Again, there is not much one can do when the
   programming information is proprietary.
   The E2100 is a poor design. Whenever it maps its
   shared memory in during a packet transfer, it
   maps it into the *whole 128K region*! That means you
   *can't* safely use another interrupt-driven shared
   memory device in that region, including another E2100.
   It will work _most_ of the time, but every once in
   a while it will bite you. (Yes, this problem can
   be avoided by turning off interrupts while
   transferring packets, but that will almost certainly
   lose clock ticks.

   Also, don't confuse the E2100 for a NE2100 clone.
   The E2100 is a shared memory NatSemi DP8390 design,
   roughly similar to a brain-damaged WD8013, whereas
   the NE2100 (and NE1500) use a bus-mastering AMD
   LANCE design.

   There is an alpha test driver available (even though
   I shouldn't have bothered) in the normal place
   (see the FAQ section) -- e2100.c -- let me know if
   you use it, and how it works. Don't forget to
   un-comment the line in config.in.

2.07 Allied Telesis

 Allied Telesis is the worlds largest maker of separate
 transceivers thanks to their low prices, and they now have a
 series of low-cost ethercards using the 79C960 version of the AMD
 LANCE. These are bus-master cards, and thus probably the fastest
 ISA bus ethercards available (although the 3c509 has lower latency
 thanks to predictive interrupts).

 Supported:

  AT1500
   The driver for the AT1500 series is new in the
   0.99pl12 kernel, but it won't work "out-of-the-box"
   with >16M machines. (NB This isn't a fundamental
   limitation, so stop pointing and laughing at the ISA
   bus. The driver just needs a hook to allocate
   low-memory buffers for the bus-master DMA, and should
   be just as fast on >16M systems. It can be easily
   fixed by initializing the LANCE driver with the
   character devices, but this fix depends on the
   resolution of the networking code uncertainty.)

   For the ISA bus master mode all structures used
   directly by the LANCE, the initialization block,
   Rx and Tx rings, and data buffers, must be accessible
   from the ISA bus, i.e. in the lower 16M of real memory.
   This is a problem for current Linux kernels on >16M
   machines. The network devices are initialized after
   memory initialization, and the kernel doles out memory
   from the top of memory downward. The current solution
   is to have a special network initialization routine
   that's called before memory initialization; this will
   eventually be generalized for all network devices.
   Low-memory "bounce-buffers" are used when needed.

   This driver should also work with NE1500 and NE2100
   clones.

   Future driver versions may figure out a way to
   autoDMA. Although there is no autoDMA (until I verify
   that autoDMA is safe and reliable), some versions
   (pl13) allow passing the DMA channel at boot-time via
   LILO. (Boot-time parameters can be made permanent in
   LILO v13+, read the docs.) The DMA channel otherwise
   defaults to DMA5.

   In pl14, there was a buglet that would hang some
   machines with AT1500 like cards. Either get pl15
   or newer, or go into ./init/main.c and move the
   sti(); and claibrate_delay(); (near line 366) in
   *front of* the #ifdef CONFIG_INET, instead of
   after it.

   Please report the exact chip used by your ethercard,
   and any success or failure you have. This driver is
   still young, and I've gotten few reports.

   More information on AMD LANCE based Ethernet cards
   can be found in section 5.08.

  AT1700
   The Allied Telesis AT1700 series ethercards are based
   on the Fujitsu MB86965. This chip uses a programmed
   I/O interface, and a pair of fixed-size transmit
   buffers. This allows small groups of packets to sent
   be sent back-to-back, with a short pause while
   switching buffers.

   A unique feature is the ability to drive 150ohm STP
   (Shielded Twisted Pair) cable commonly installed for
   Token Ring, in addition to 10baseT 100ohm UTP
   (unshielded twisted pair).

   A mis-feature to watch out for is that the current
   production version silently wires to DMA channel 5,
   rendering it useless. No device driver will be
   written using DMA if installing a second card into
   the machine breaks both, and the only way to disable
   the DMA is with a knife.

   The at1700 driver is included in the standard pl15
   kernel source tree.

2.08 Arcnet

 There is no Arcnet driver for Linux. Feel free to write a driver. With
 the very low cost and better performance of ethernet, I expect that
 most places will be giving away their Arcnet hardware for free,
 resulting in a lot of home systems with Arcnet.

 An advantage of Arcnet is that all of the cards have identical
 interfaces, so once a driver is available it will work for everyone.

2.09 Digital / DEC

 Supported: DE200, DE210, DE202, DE100, DEPCA rev E

  As of linux v1.0, there is a driver included as standard
  for these cards. It was written by David C. Davies.
  There is documentation included in the source file
  "depca.c", which includes info on how to use more than
  one of these cards in a machine.

  If you have / want to use the pl15 kernel or older,
  then you will have to use Peter Bauer's driver.
  It can be found as a separate patch called depca-0.8.tar.gz.
  You will have to un-comment the DEPCA line in "config.in"
  after installing the patch. You can find the patch on
  ftp.funet.fi, /pub/OS/Linux/BETA/depca/depca-0.8.tar.gz
  This version resets the card upon close so that you can
  use it with broken DOS drivers after a warm boot.


 Unsupported: Digital Etherlink III

  Peter Bauer said that "the new etherlink III seems to
  be a break: No official docu from DEC as far as today,
  other (incompatible??) hardware used, and (no joke) (at least
  for the first delivered cards) also a sharp knife necessary
  to get the card working (needs cut of some irq lines ...)
  As far as I know, lots of DEC Employees use Linux (at least
  for hobby purposes) and the depca-driver, because its a
  de-facto standard in DEC, so I encourage any DEC-employee
  reading this to check wether my writing is true, and to
  support sources of information about the etherworks-III."

2.10 Intel Ethernet Cards

 Supported: Ether Express

  This card uses the intel i82586. (Surprise, huh?)
  The driver is in the standard release of pl15.
  However, you will have to uncomment the line in
  "config.in" to use it. -- yes, this line is
  commented out for a reason. The driver is still
  in the testing phases, as of v1.0 as well.

  There is some technical information available on
  the i82586 in section 5.06, and also in the
  source code for the driver "eexpress.c". Don't
  be afraid to read it. ;-)

  The rason is that the driver works well with slow machines,
  but the i82586 occasionally hangs from the packet buffer
  contention that a fast machine can cause. I'll have
  to find a work-around before releasing the driver.
  One reported hack fix is to change all of the outw()
  calls to outw_p().


  If you do try the driver please post or send a report.
  Include the kind of machine you are trying it with,
  and how heavily loaded your network is.


2.11 PureData

 Supported: PDUC8028, PDI8023

  The PureData PDUC8028 and PDI8023 series of cards are reported
  to work, thanks to special probe code contributed by Mike
  Jagdis <[e-mail]>. The support is integrated
  with the WD driver.

2.12 Xircom

 Another group that won't release documentation. No cards
 supported. Don't look for any support in the future unless
 they release their programming information. And this is
 highly unlikely, as they *forbid* you from even reverse-
 engineering their drivers. If you are already stuck with one,
 see if you can trade it off on some DOS (l)user. Read section
 9.01 if you are bored.

 And if you just want to verify that this is the case, you can
 reach Xircom at 1-800-874-7875 or [phone].

2.13 Zenith

 The built-in Z-Note network adaptor is based on the Intel
 i82593 using two DMA channels. There might be a driver for it
 in mid 1994. See section 5.06 for more information.
 Also note that the Z-Note is compatible with the IBM ThinkPad 300.

2.14 Racal-Interlan

 Note: I have been told that the following two drivers are
 for patchlevel 11, and hence are a bit dated. The original
 author is Michael Hipp, and can be reached at the following addr:
  [e-mail]

 NI52**

 There is an alpha driver for the NI5210 floating about.
 (last seen on tsx-11.mit.edu /pub/linux/ALPHA/ni/ni52.tar.gz)
 This card also uses one of the Intel chips. See section
 5.06 for more information.

 NI65**

 There is also a driver for the LANCE based NI6510, and it
 can be found in the same place as the NI5210 driver above.
 I am not sure how much work it would be to hack the current
 LANCE driver in the kernel to support this card. If anyone
 has done so, let me know.

2.15 AMD LANCE (79C960)

 There really is no AMD ethernet card. You are probably reading this
 because the only markings you could find on your card said AMD
 and the above number. The above number refers to a chip from AMD
 that is the heart of many ethernet cards. See the section on the
 Allied Telesis AT1500, the NE1500/2100 and the information in
 section 5.08. Chances are that the existing LANCE driver will work
 with all AMD LANCE based cards. (...except perhaps the above
 mentioned NI6510 ???)

2.16 AT-Lan-Tec / RealTek Pocket adaptor

 This is a generic, low-cost OEM pocket adaptor being sold by
 AT-Lan-Tec, and (likely) a number of other suppliers. A
 driver for it is included in the standard pl15 kernel.
 Note that there is substantial information contained in the
 driver source file "atp.c" which presently lives in ./drivers/net/
 BTW, the adaptor (AEP-100L) has both 10baseT and BNC connections!
 You can reach AT-Lan-Tec at 1-301-948-7070. Ask for the model
 that works with Linux, or ask for "Vincent Bono" in tech support.
 In the Netherlands a compatible adaptor is sold under the name SHI-TEC
 PE-NET/CT, and sells for about $125. The vendor was Megasellers.
 They state that they do not sell to private persons, but I just
 gave them the name of my home institute. No questions asked.
 They are: Megasellers, Vianen, The Netherlands. They always
 advertise in Dutch computer magazines. In Germany, a similar
 adaptor comes as a no-brand-name product. Prolan 890b, no
 brand on the casing, only a roman II. Resellers can get a price
 of about $130, including a small wall transformer for the power.

 Physical Description

 The adaptor is "normal size" for the product class, about 57mm wide,
 22mm high tapering to 15mm high at the DB25 connector, and 105mm long
 (120mm including the BNC socket). It's switchable between the RJ45
 and BNC jacks with a small slide switch positioned between the two:
 a very intuitive design.

 It's powered by a lightweight 5V "wall brick" adaptor that terminates
 in a standard 5.0mm power connector. I measured an unconnected
 quiescent power draw of 102ma for BNC and 84ma for 10baseT. I hooked
 the pocket adaptor up to my home thinnet and started FTPing a large
 file. The power measurements were:

  idle, connected  99ma @ 5.1V
  active, connected 107ma @ 5.1V

 This was measured using a Fluke 8026B true-RMS multimeter, so I'm
 pretty confident the numbers are good. This power draw is low enough
 that you could buy or build a cable to take the 5V directly from the
 keyboard/mouse port available on many laptops. (Bonus points here
 for using a standardized power connector instead of proprietary one.)

2.17 Ansel

 Supported: AC3200 EISA

  This driver is not included in the pl15 kernel. To
  *alpha* test it, get the files ac3200.[c,h] from
  where you usually get alpha drivers (see the FAQ in
  this document if you dont know) and uncomment the
  line in config.in for the ac3200. If you use it,
  please let me know how things work out.

2.18 DFI

 Supported: DFINET-300 (NE1000) and DFINET-400 (NE2000)

  These cards are now detected (as of pl15) thanks to
  Eberhard Moenkeberg <[e-mail]> who noted that
  they use "DFI" in the first 3 bytes of the prom, instead
  of using 0x57 in bytes 14 and 15, which is what all the
  NE1000 and NE2000 cards use.

3. Clones of popular Ethernet cards.

 Due to the popular design of some cards, different companies will
 make "clones" or replicas of the original card. However, one must
 be careful, as some of these clones are not 100% compatible, and
 can be troublesome. Some common problems with "not-quite-clones"
 are noted in the question and answer section of this document.

 Also note that if your card isn't mentioned here, that really
 means nothing. Chances are that even if it is only a half decent
 clone of the original, then it will still work.

3.1 WD80x3 clones

 The following clones are reported to work with the standard
 WD80x3 driver:

 AT-LAN-TEC 8013
 PureData (not a 8013 clone, but the 8013 driver has special code)
 LANNET LEC-45
 PE-8013 (WD-8013 Compatible)

3.2 NE2000 clones

 The following clones are reported to work with the standard
 NE2000 driver:

 Accton NE2000 (might not get detected at boot, see section 6)
 Alta Combo NE2000 clone
 Aritsoft LANtastic AE-2 (OK, but has flawed error-reporting registers)
 Asante Etherpak 2001/2003
 AT-LAN-TEC NE2000 clone (uses Winbond chip that traps SCSI drivers)
 Cabletron products: E10**, E10**-x, E20**, E20**-x
 Cnet UTP 10baseT (NE 2000 emulation)
 D-Link Ethernet II (bad clones, but the driver checks for them)
 4-Dimension FD0490 EtherBoard16
 LTC E-NET/16 P/N: 8300-200-002 ([e-mail])
 Network Solutions HE-203
 SIIG Inc E-Lan/200 (NE 2000 comp.)
 SVEC 4 Dimension Ethernet

4. Cables, coax, twisted pairs etc.

 If you are starting a network from scratch, it's considerably less
 expensive to use thin ethernet, RG58 co-ax cable with BNC connectors,
 than old-fashioned thick ethernet, RG-5 cable with N connectors, or
 10baseT, twisted pair telco-style cables with RJ-45 eight wire "phone"
 connectors.

4.01 Thin Ethernet (thinnet)

 Thin ethernet is the "ether of choice". The cable is inexpensive. If
 you are making your own cables solid-core RG58A is $0.09/ft. and
 stranded RG58AU is $0.15/ft. Twist-on BNC connectors are < $2 ea.,
 and other misc. pieces are similarly inexpensive. It is essential
 that you properly terminate each end of the cable with 50 ohm
 terminators, so budget $2 ea. for a pair. It's also vital that
 your cable have no "stubs" -- the 'T' connectors must be attached
 directly to the ethercards. The only drawback is that if you have
 a big loop of machines connected together, and some bonehead breaks
 the loop by taking one cable off the side of his tee, the whole
 network goes down because it sees an infinite impedance (open
 circuit) instead of the required 50 ohm termination. Note that
 you can remove the tee piece from the card itself without killing
 the whole subnet, as long as you don't remove the cables from the
 tee itself. Of course this will disturb the machine that you
 pull the actual tee off of. 8-) And if you are doing a small
 network of two machines, you *still* need the tees and the 50 ohm
 terminators -- you *can't* just cable them together!


4.02 Twisted pair

 Twisted pair networks require active hubs, which start around $200,
 and the raw cable cost can actually be higher than thinnet. They are
 usually sold using the claim that you can use your existing telephone
 wiring, but it's a rare installation where that turns out to be the
 case. The claim that you can upgrade to higher speeds is also
 suspect, as most proposed schemes use higher-grade (read $$) cable and
 more sophisticated termination ($$$) than you would likely install on
 speculation. New gizmos are floating around which allow you to
 daisy-chain machines together, and the like. For example,
 Falleron sells EtherWave adaptors and transceivers. This device
 allows multiple 10baseT devices to be daisy-chained. They also
 sell a 3c509 clone that includes the EtherWave transceiver.
 The drawback is that it's more expensive and less reliable than a
 cheap ($100-$150) mini-hub and another ethercard. IMO, you should
 either go for the hub approach or switch over to 10base2 thinnet.

 On the other hand, hubs are rapidly dropping in price, all 100Mb/sec
 ethernet proposals use twisted pair, and most new business
 installations use twisted pair. (This is probably to avoid the
 problem with idiots messing with the BNC's as described above.)

 If you are only connecting two machines, it is possible to avoid
 using a hub, by swapping the Rx and Tx pairs (1-2 and 3-6).

 Also, Russ Nelson adds that "New installations should use Category 5
 wiring. Anything else is a waste of your installer's time, as
 100Base-whatever is going to require Cat 5."

4.03 Thick Ethernet

 Thick ethernet is mostly obsolete, and is usually used only to remain
 compatible with an existing implementation. You can stretch the rules
 and connect short spans of thick and thin ethernet together with a
 passive $3 N-to-BNC connector, and that's often the best solution to
 expanding an existing thicknet. A correct (but expensive) solution is
 to use a repeater in this case.


5 Technical information.

 For those who want to play with the present drivers, or try to make
 up their own driver for a card that is presently unsupported, this
 information should be useful. If you do not fall into this category,
 then perhaps you will want to skip this section.

5.01 Probed addresses

 While trying to determine what ethernet card is there, the following
 addresses are autoprobed, assuming the type and specs of the card
 have not been set in the kernel. As of 0.99pl12, doing a "make config"
 will ask what cards are to be supported. The file names below are
 in /usr/src/linux/drivers/net/
 ----------------------------------------------------------------
 wd.c:  0x300, 0x280, 0x380, 0x240
 3c501.c  0x280
 3c503.c: 0x300, 0x310, 0x330, 0x350, 0x250, 0x280, 0x2a0, 0x2e0
 3c507.c: 0x300, 0x320, 0x340, 0x280
 3c509.c: <Special "ID Port" probe>
 at1700.c: 0x300, 0x280, 0x380, 0x320, 0x340, 0x260, 0x2a0, 0x240
 atp.c:  0x378, 0x278, 0x3bc
 depca.c  0x300, 0x200
 d_link.c: 0x378
 ne.c:  0x300, 0x280, 0x320, 0x340, 0x360
 hp.c:  0x300, 0x320, 0x340, 0x280, 0x2C0, 0x200, 0x240
 lance.c: 0x300, 0x320, 0x340, 0x360
 smc-ultra.c: 0x200, 0x220, 0x240, 0x280, 0x300, 0x340, 0x380
 eexpress.c: 0x300, 0x270, 0x320, 0x340
 3c509.c: <Special "ID Port" probe>
 ----------------------------------------------------------------
 There are some NE2000 clone ethercards out there that are waiting black
 holes for autoprobe drivers. While many NE2000 clones are
 safe until they are enabled, some can't be reset to a safe mode.
 These dangerous ethercards will hang any I/O access to their
 "dataports". The typical dangerous locations are:

 Ethercard jumpered base     Dangerous locations (base + 0x10 - 0x1f)
  0x300 *    0x310-0x317
  0x320    0x330-0x337
  0x340    0x350-0x357
  0x360    0x370-0x377

 * The 0x300 location is the traditional place to put an ethercard, but
 it's also a popular place to put other devices (often SCSI
 controllers). The 0x320 location is often the next one chosen, but
 that's bad for for the AHA1542 driver probe. The 0x360 location is
 bad, because it conflicts with the parallel port at 0x378.

 To avoid these lurking ethercard, here are the things you can do:

 o Probe for the device's BIOS in memory space. This is easy
   and always safe, but it only works for cards that always have
   BIOSes, like primary SCSI controllers.

 o Avoid probing any of the above locations until you think
   you've located your device. The NE2000 clones have a reset range
   from <base>+0x18 to <base>+0x1f that will read as 0xff, so probe
   there first if possible. It's also safe to probe in the 8390
   space at <base>+0x00 - <base>+0x0f, but that area will return
   quasi-random values

 o If you must probe in the dangerous range, for instance if your
   target device has only a few port locations, first check that
   there isn't an NE2000 there. You can see how to do this by
   looking at the probe code in /usr/src/linux/net/inet/ne.c

 In other news, I've written the code for the I/O port registrar.
 Peter MacDonald and I have been intensely discussing this, and I think
 our current scheme has the necessary functionality with minimal kernel
 size impact. (The implementation involved rewriting the bitmap ops in
 kernel/ioport.c:ioperm() so that most code could be shared.)

 Here is the current "blurb". As usual comments are welcome. Please
 keep them substantial and constructive (we've already talked about
 changing the name from "reserve=" to "noprobe=").

 ==================

 Boot-Time Parameters: "reserve="

 In some machines it may be necessary to prevent device drivers from
 checking for devices (auto-probing) in a specific region. This may be
 because of poorly designed hardware that causes the boot to "freeze"
 (such as some ethercards), hardware that is mistakenly identified,
 hardware whose state is changed by an earlier probe, or merely
 hardware you don't want the kernel to initialize.

 The "reserve" boot-time argument addresses this problem by specifying
 an I/O port region that shouldn't be probed. That region is reserved
 in the kernel's port registration table as if a device has already
 been found in that region. Note that this mechanism shouldn't be
 necessary on most machine, only when there is a problem or special
 case.

 The boot-line syntax is

   lilo-prompt: linux-image reserve=[<port>,<size>,<port>,<size>...]

 As usual with boot-time specifiers there is an 11 parameter limit, thus
 you can only specify 5 reserved regions per "reserve" keyword.
 Multiple "reserve" specifiers will work if you have an usually
 complicated request.

 If you specify a "reserve" region to protect a specific device, you
 must generally specify an explicit probe for that device. Most
 drivers ignore the port registration table if they are given an
 explicit address.

5.02 Skeleton / prototype driver

 OK. So you have decided that you want to write a driver for the
 Foobar Ethernet card, as you have the programming information,
 and it hasn't been done yet. (...these are the two main require-
 ments ;-) You can use the skeleton network driver that is provided
 with the Linux kernel source tree. It can be found in the file
 /usr/src/linux/drivers/net/skeleton.c as of 0.99pl15, and later.

 It's also very useful to look at the Crynwr (nee Clarkson) driver
 for your target ethercard, if it's available. Russ Nelson
 <[e-mail]> has been actively updating and writing these,
 and he has been very helpful with his code reviews of the current
 Linux drivers.

5.03 Driver interface to the kernel

 Here are some notes that may help when trying to figure out what
 the code in the driver segments is doing, or perhaps what it is
 supposed to be doing.

 =====================================================

 int ethif_init(struct device *dev)
 {
     ...
  dev->send_packet = &ei_send_packet;
  dev->open = &ei_open;
  dev->stop = &ei_close;
  dev->hard_start_xmit = &ei_start_xmit;
  ...
 }

 int ethif_init(struct device *dev)

 This function is put into the device structure in Space.c. It is
 called only at boot time, and returns '0' iff the ethercard 'dev'
 exists.

 =====================================================

 static int ei_open(struct device *dev)
 static int ei_close(struct device *dev)

 This routine opens and initializes the board in response to an
 socket ioctl() usually called by 'config' or 'ifconfig'. It is
 commonly stuffed into the 'struct device' by ethif_init().

 The inverse routine is ei_close(), which should shut down the
 ethercard, free the IRQs and DMA channels if the hardware permits,
 and turn off anything that will save power (like the transceiver).

 (Note: As of NET-2, the relevant program is '/etc/ifconfig' - and
 the device *can* be turned off or on via passing 'up' or 'down'
 to 'ifconfig' from the command line with the device name.)

 =====================================================

 static int ei_start_xmit(struct sk_buff *skb, struct device *dev)
  dev->hard_start_xmit = &ei_start_xmit;

 This routine puts packets to be transmitted into the hardware. It
 is usually stuffed into the 'struct device' by ethif_init().

 When the hardware can't accept additional packets it should set
 the dev->tbusy flag. When additional room is available, usually
 during a transmit-complete interrupt, dev->tbusy should be cleared
 and the higher levels informed with mark_bh(INET_BH).
 [[Note: pre0.99.4 kernels didn't use this interface for all packets.]]

 =====================================================

 ...
     if (dev_rint(buffer, length, is_skb ? IN_SKBUFF : 0, dev))
     stats->rx_dropped++;
 ...
 A received packet is passed to the higher levels using dev_rint().
 If the unadorned packet data in a memory buffer, dev_rint will copy
 it into a 'skbuff' for you. Otherwise a new skbuff should be
 kmalloc()ed, filled, and passed to dev_rint() with the IN_SKBUFF flag.

 =====================================================

5.04 Interrupts and Linux

 There are two kinds of interrupt handlers in Linux:
 fast ones and slow ones. You decide what kind you are installing by
 the flags you pass to irqaction(). The fast ones, such as the serial
 interrupt handler, run with _all_ interrupts disabled. The normal
 interrupt handlers, such as the one for ethercard drivers, runs with
 other interrupts enabled.

 There is a two-level interrupt structure. The "fast" part handles the
 device register, removes the packets, and perhaps sets a flag.  After
 it is done, and interrupts are re-enabled, the slow part is run if the
 flag is set.

 The flag between the two parts is set by:
  mark_bh(INET_BH);

 Usually this flag is set within dev_rint() during a received-packet
 interrupt, and set directly by the device driver during a
 transmit-complete interrupt.

 You might wonder why all interrupt handlers cannot run in
 "normal mode" with other interrupts enabled. Ross Biro uses this
 scenario to illustrate the problem:
  o You get a serial interrupt, and start processing it.
    The serial interrupt is now masked.
  o You get a network interrupt, and you start transferring
    a maximum-sized 1500 byte packet from the card.
  o Another character comes in, but this time the interrupts
    are masked!

 The "fast" interrupt structure solves this problem by allowing
 bounded-time interrupt handlers to run without the risk of leaving
 their interrupt lines masked by another interrupt request.

 There is an additional distinction between fast and slow interrupt
 handlers -- the arguments passed to the handler. A "slow" handler is
 defined as

  static void
  handle_interrupt(int reg_ptr)
  {
      int irq = -(((struct pt_regs *)reg_ptr)->orig_eax+2);
      struct device *dev = irq2dev_map[irq];
  ...

 While a fast handler gets the interrupt number directly

  static void
  handle_fast_interrupt(int irq)
  {
  ...

 A final aspect of network performance is latency. The only board
 that really addresses this is the 3c509, which allows a predictive
 interrupt to be posted. It provides an interrupt response timer so
 that the driver can fine-tune how early an interrupt is generated.

 Alan Cox has some advice for anyone wanting to write drivers
 that are to be used with pl14 kernels and newer. He says:

 "Any driver intended for pl14 should use the new alloc_skb() and
 kfree_skbmem() functions rather than using kmalloc() to obtain an
 sk_buff. The new pl14 skeleton does this correctly. For drivers
 wishing to remain compatible with both sets the define
 'HAVE_ALLOC_SKB' indicates these functions must be used.

 In essence replace

  skb=(struct sk_buff *)kmalloc(size)
 with

  skb=alloc_skb(size)

 and

  kfree_s(skb,size)

 with

  kfree_skbmem(skb,size) /* Only sk_buff memory though */

 Any questions should I guess be directed to me since I made the change.
 This is a change to allow tracking of sk_buff's and sanity checks on
 buffers and stack behaviour. If a driver produces the message
 'File: ??? Line: ??? passed a non skb!' then it is probable the
 driver is not using the new sk_buff allocators."


5.05 Programmed I/O vs. shared mem. vs. slave/master DMA

 Ethernet is 10Mbs. (Don't be pedantic, 3Mbs and 100Mbs don't count.)
 If you can already send and receive back-to-back packets, you just
 can't put more bits over the wire. Every modern ethercard can receive
 back-to-back packets. The Linux DP8390 drivers come pretty close to
 sending back-to-back packets (depending on the current interrupt
 latency) and the 3c509 and AT1500 hardware has no problem at all
 automatically sending back-to-back packets.

 The ISA bus can do 5.3MB/sec (42Mb/sec), which sounds like more than
 enough. You can use that bandwidth in several ways:

 Programmed I/O
 ==============
   Pro: Doesn't use any constrained system resources,
        just a few I/O registers, and has no 16M limit.
   Con: Usually the slowest transfer rate, the CPU is waiting
        the whole time, and interleaved packet access is usually
        difficult to impossible.

 Shared memory
 =============
   Pro: Simple, faster than programmed I/O, and allows random
        access to packets.
   Con: Uses up memory space (a big one for DOS users, only a minor
        issue under Linux), and it still ties up the CPU.

 Slave (normal) DMA
 ==================
   Pro: Frees up the CPU during the actual data transfer.
   Con: Checking boundary conditions, allocating contiguous buffers,
        and programming the DMA registers makes it the slowest
        of all techniques.  It also uses up a scarce DMA
        channel, and requires aligned low memory buffers.

 Master (bus-master) DMA
 =======================
   Pro: Frees up the CPU during the data transfer, can string together
        buffers, can require little or no CPU time lost on the
        ISA bus.
   Con: Requires low-memory buffers and a DMA channel. Any
        bus-master will have problems with other bus-masters that
        are bus-hogs, such as some primitive SCSI adaptors. A few
        badly-designed motherboard chipsets have problems with
        bus-masters. And a reason for not using *any* type of
        DMA device is using a Cyrix 486 processor designed for
        plug-in replacement of a 386: these processors must
        flush their cache with each DMA cycle.

---End of part 1/2---