Collection of Internet/Collection of Internet.iso/faq/comp/linux/howto/ethernet/part2

Internet Message Format | 1994-03-15 | 52KB

Path: bloom-beacon.mit.edu!hookup!swrinde!emory!news-feed-2.peachnet.edu!concert!bigblue.oit.unc.edu!sunSITE!mdw
From: Paul Gortmaker <[e-mail]>
Newsgroups: comp.os.linux.announce,comp.os.linux.admin,comp.answers,news.answers
Subject: Linux Ethernet HOWTO (Part 2/2)
Followup-To: poster
Date: 16 Mar 1994 05:27:09 GMT
Organization: The University of North Carolina at Chapel Hill
Lines: 1178
Approved: [e-mail] (Matt Welsh)
Message-ID: <2m65bd$[e-mail]>
NNTP-Posting-Host: calzone.oit.unc.edu
Keywords: Linux, Ethernet, TCP/IP, NET-2, network
Originator: mdw@sunSITE
Xref: bloom-beacon.mit.edu comp.os.linux.announce:1952 comp.os.linux.admin:5613 comp.answers:4182 news.answers:16451

Archive-Name: linux/howto/ethernet/part2
Last-Modified: 16 Mar 1994

[This is part 2/2 of the Ethernet-HOWTO.]

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.

5.06 Programming the Intel chips (i82586 and i82593)

    These chips are used on a number of cards, namely the 3c507 ('86),
    the Intel EtherExpress 16 ('86), Microdyne's exos205t ('86),
    the Z-Note ('93), and the Racal-Interlan ni5210 ('86).

    Russ Nelson writes:
    "Most boards based on the 82586 can reuse quite a bit of their code.
    More, in fact, than the 8390-based adapters. There are only three
    differences between them:

      o The code to get the Ethernet address,
      o The code to trigger CA on the 82586, and
      o The code to reset the 82586.

    The Intel EtherExpress 16 is an exception, as it I/O maps the 82586.
    Yes, I/O maps it. Fairly clunky, but it works.

    Garrett Wollman did an AT&T driver for BSD that uses the BSD
    copyright. The latest version I have (Sep '92) only uses a single
    transmit buffer. You can and should do better than this if you've
    got the memory. The AT&T and 3c507 adapters do; the ni5210 doesn't.

    The people at Intel gave me a very big clue on how you queue up
    multiple transmit packets. You set up a list of
    NOP->XMIT->NOP->XMIT->NOP->XMIT->(beginning) blocks, then you set the
    "next" pointer of all the NOP blocks to themselves. Now you start
    the command unit on this chain. It continually processes the first
    NOP block. To transmit a packet, you stuff it into the next transmit
    block, then point the NOP to it. To transmit the next packet, you
    stuff the next transmit block and point the previous NOP to *it*. In
    this way, you don't have to wait for the previous transmit to finish,
    you can queue up multiple packets without any ambiguity as to whether
    it got accepted, and you can avoid the command unit start-up delay."

5.07    Technical information from 3Com

    From: Cameron Spitzer 764-6339 <[e-mail]>
    Subject: getting 3Com Adapter manuals
    Date: Mon, 27 Sep 1993 21:17:07 +0200

    Since this is becoming a FAQ, I'm going to tread the thin
    ice of No Commercial Use and answer it here.

    3Com's Ethernet Adapters are documented for driver writers
    in our "Technical References" (TRs). These manuals describe
    the programmer interfaces to the boards but they don't talk
    about the diagnostics, installation programs, etc that end
    users can see.

    The Network Adapter Division marketing department has the
    TRs to give away. To keep this program efficient, we
    centralized it in a thing called "CardFacts." CardFacts is
    an automated phone system. You call it with a touch-tone
    phone and it faxes you stuff. To get a TR, call CardFacts
    at 408-727-7021. Ask it for Developer's Order Form,
    document number 9070. Have your fax number ready when you
    call. Fill out the order form and fax it to 408-764-5004.
    Manuals are shipped by Federal Express 2nd Day Service.

    If you don't have a fax and nobody you know has a fax,
    really and truly, *then* send mail to
    [e-mail] and tell her about your problem.
    PLEASE use the fax thing if you possibly can.

    After you get a manual, if you still can't figure out how to
    program the board, try our "CardBoard" BBS at
    1-800-876-3266, and if you can't do that, write
    [e-mail] and ask him for alternatives. If
    you have a real stumper that nobody has figured out yet, the
    fellow who needs to know about it is
    [e-mail].

    There are people here who think we are too free with the
    manuals, and they are looking for evidence that the system
    is too expensive, or takes too much time and effort. That's
    why it's important to try to use CardFacts *before* you
    start calling and mailing the people I named here.

    There are even people who think we should be like Diamond
    and Xircom, requiring tight "partnership" with driver
    writers to prevent poorly performing drivers from getting
    written. So far, 3Com customers have been really good about
    this, and there's no problem with the level of requests
    we've been getting. We need your continued cooperation and
    restraint to keep it that way.

    Cameron Spitzer, 408-764-6339
    3Com NAD
    Santa Clara
    work: [e-mail]
    home: [e-mail]

5.08 Notes on AMD PCnet-ISA / LANCE Based cards (79C960)

    The AMD LANCE (Local Area Network Controller for Ethernet)
    was the original offering, and has since been replaced by
    the "PCnet-ISA" chip, otherwise known as the 79C960.
    A relatively new chip from AMD, the 79C960, is the heart of many
    new cards being released at present. Note that the name "LANCE"
    has stuck, and some people will refer to the new chip by the old
    name. Dave Roberts of the Network Products Division of AMD was kind
    enough to contribute the following information regarding this chip:

    "As for the architecture itself, AMD developed it originally
    and reduced it to a single chip -- the PCnet(tm)-ISA -- over a year
    ago. It's been selling like hotcakes ever since.

    Functionally, it is equivalent to a NE1500. The register set
    is identical to the old LANCE with the 1500/2100 architecture
    additions. Older 1500/2100 drivers will work on the PCnet-ISA.
    The NE1500 and NE2100 architecture is basically the same.
    Initially Novell called it the 2100, but then tried to distinguish
    between coax and 10BASE-T cards. Anything that was 10BASE-T only was
    to be numbered in the 1500 range. That's the only difference.

    Many companies offer PCnet-ISA based products, including HP,
    Racal-Datacom, Allied Telesis, Boca Research, Kingston Technology, etc.
    The cards are basically the same except that some manufacturers
    have added "jumperless" features that allow the card to
    be configured in software. Most have not. AMD offers a standard
    design package for a card that uses the PCnet-ISA and many
    manufacturers use our design without change.
    What this means is that anybody who wants to write drivers for
    most PCnet-ISA based cards can just get the data-sheet from AMD. Call
    our literature distribution center at [phone] and ask for the
    Am79C960, PCnet-ISA data sheet. It's free.

    A quick way to understand whether the card is a "stock" card
    is to just look at it. If it's stock, it should just have one large
    chip on it, a crystal, a small IEEE address PROM, possibly a socket
    for a boot ROM, and a connector (1, 2, or 3, depending on the media
    options offered). Note that if it's a coax card, it will have some
    transceiver stuff built onto it as well, but that should be near the
    connector and away from the PCnet-ISA.

    The PCnet-ISA is faster than the original LANCE design and
    makes better use of the available bus bandwidth. Additionally, some
    LANCE bugs were corrected and many enhancements were made."

    There is also some info regarding the LANCE chip in the file
    lance.c which is included in the standard kernel.

5.09 Multicast and Promiscuous mode

    One of the things I've been working on recently is the
    major remaining item on the ethercard feature list:
    implementing multicast and promiscuous mode hooks.

    At first I was planning to do it while implementing either
    the /dev/* or DDI interface, but that's not really the
    correct way to do it. We should only enable multicast or
    promiscuous modes when something wants to look at the
    packets, and shut it down when that application is
    finished, neither of which is strongly related to when the
    hardware is opened or released.

    I'll start by discussing promiscuous mode, which is
    conceptually easy to implement. For most hardware you
    only have to set a register bit, and from then on you get
    every packet on the wire. Well, it's almost that easy;
    for some hardware you have to shut the board (potentially
    dropping a few packet), reconfigure it, and then re-enable
    the ethercard. This is grungy and risky, but the
    alternative seems to be to have every application register
    before you open the ethercard at boot-time.

    OK, so that's easy, so I'll move on something that's not
    quite so obvious: Multicast. It can be done two ways:

    1) Use promiscuous mode, and a packet filter like the
       Berkeley packet filter (BPF). The BPF is a pattern matching
       stack language, where you write a program that picks out the
       addresses you are interested in. Its advantage is that it's
       very general and programmable. Its disadvantage is that there
       is no general way for the kernel to avoid turning on promiscuous
       mode and running every packet on the wire through every registered
       packet filter. See the next section for more information on BPF.

    2) Using the built-in multicast filter that most etherchips have.

    I guess I should list what a few ethercards/chips provide:

    Chip/card  Promiscuous    Multicast filter
    ========================================
    Seeq8001/3c501    Yes    Binary filter (1)
    3Com/3c509    Yes    Binary filter (1)
    8390        Yes    Autodin II six bit hash (2) (3)
    LANCE        Yes    Autodin II six bit hash (2) (3)
    i82586        Yes    Hidden Autodin II six bit hash (2) (4)


    (1) These cards claim to have a filter, but it's a simple
    yes/no 'accept all multicast packets', or 'accept no
    multicast packets'.

    (2) AUTODIN II is the standard ethernet CRC (checksum)
    polynomial. In this scheme multicast addresses are hashed
    and looked up in a hash table. If the corresponding bit
    is enabled, this packet is accepted. Ethernet packets are
    laid out so that the hardware to do this is trivial -- you
    just latch six (usually) bits from the CRC circuit (needed
    anyway for error checking) after the first six octets (the
    destination address), and use them as an index into the
    hash table (six bits == a 64-bit table).

    (3) These chips use the six bit hash, and must have the
    table computed and loaded by the host. This means the
    kernel must include the CRC code.

    (4) The 82586 uses the six bit hash internally, but it
    computes the hash table itself from a list of multicast
    addresses to accept.

    Note that none of these chips do perfect filtering, and we
    still need a middle-level module to do the final
    filtering. Also note that in every case we must keep a
    complete list of accepted multicast addresses to recompute
    the hash table when it changes.

    My first pass at device-level support is detailed in the
    new outline driver skeleton.c (pl14 and up.)

    It looks like the following:

    #ifdef HAVE_MULTICAST
    static void set_multicast_list(struct device *dev, int num_addrs,
                     void *addrs);
    #endif
    .
    .

    ethercard_open() {
    ...
#ifdef HAVE_MULTICAST
        dev->set_multicast_list = &set_multicast_list;
    #endif
    ...

    #ifdef HAVE_MULTICAST
    /* Set or clear the multicast filter for this adaptor.
       num_addrs == -1    Promiscuous mode, receive all packets
       num_addrs == 0    Normal mode, clear multicast list
       num_addrs > 0    Multicast mode, receive normal and
                MC packets, and do best-effort filtering.
     */
    static void
    set_multicast_list(struct device *dev, int num_addrs, void *addrs)
    {
    ...

    Any comments, criticism, etc. are welcome.

    Alan Cox adds that "...in pl14, user programs can access promiscuous
    mode but not multicast mode, even though the drivers support both.
    The ifconfig program allows you to mark an interface 'promisc'."

5.10 The Berkeley Packet Filter (BPF)

    I'm not bitterly opposed to it, but I'm coming to the
    conclusion that the 'bpf' functionality should not be provided
    by the kernel, but should be in a (hopefully little-used)
    compatibility library.

    For those not in the know: 'bpf' (the Berkeley Packet Filter)
    is an mechanism for specifying to the kernel networking layers
    what packets you are interested in. It's implemented as a
    specialized stack language interpreter built into a low level
    of the networking code. An application passes a program
    written in this language to the kernel, and the kernel runs the
    program on each incoming packet. If the kernel has multiple
    'bpf' applications, each program is run on each packet.

    The problem is that it's difficult to deduce what kind of
    packets the application is really interested in from the packet
    filter program, so the general solution is to always run the
    filter. Imagine a program that registers a 'bpf' program to
    pick up a low data-rate stream sent to a multicast address.
    Most ethernet cards have a hardware multicast address filter
    implemented as a 64 entry hash table that ignores most unwanted
    multicast packets, so the capability exists to make this a very
    inexpensive operation. But with the BFP the kernel must switch
    the interface to promiscuous mode, receive _all_ packets, and
    run them through this filter. This is work, BTW, that's very
    difficult to account back to the process requesting the packets.

5.11 Unresolved questions / concerns

    There may be some benefit from processing packet data as it is
    transferred to and from the ethercard, especially with very fast
    processors transferring data to a slow ethercard. As I see it this
    question has multiple parts:
        1) Is there any useful processing power available, perhaps
           during the ISA bus recovery period, or while the 8390
           remote DMA is preparing for another transfer??
        2) Is there any useful but simple work that can be done
           between/during each word of the copy, such as calculating
           a CRC, or discarding obviously unwanted packets??
        3) would the complexity of an interface to do this make future
           ethercard drivers impossible??

    There should be a better structure than Space.c - Drivers should be
    able to autoprobe for all installed ethercards rather than just
    quitting after finding the first.  I've written code to do this,
    but the constant promise (threat?) of DDI has prevented me from
    making it standard.

    A related topic is the problem of driver probes corrupting
    unrelated hardware. Even worse is a probe into a dataport that
    isn't set up to transfer data, which will freeze the machine. The
    common suggestion is a boot-time device registry that records
    already-used I/O ports and shared memory. This has been implemented
    as of pl13, see section 5.01.

6 Possible problems, and troubleshooting.

    This section tries to answer any unresolved questions, and not so
    common solutions to common problems. They are sorted on a "per
    manufacturer basis". You should have also read the relevant info.
    from section 1 about your specific card. Section 8 contains more
    general FAQ's.

6.01 Problems with NE2000 (and clones)

    "DMA address mismatch"
    ======================

    Is the chip a real NatSemi 8390? (DP8390, DP83901, DP83902 or DP83905)?
    If not, some clone chips don't correctly implement the transfer
    verification register. MS-DOS drivers never do error checking,
    so it doesn't matter to them.

    Are most of the messages off by a factor of 2?
    If so:    Are you using the NE2000 in a 16 bit slot?
        Is it jumpered to use only 8 bit transfers?

    The Linux driver expects a NE2000 to be a 16 bit slot. A NE1000 can
    be in either size slot. This problem can also occur with some clones,
    notably D-Link 16 bit cards, that don't have the correct ID bytes
    in the station address PROM. [[ This should be fixed in pl12.]]

    Are you running the bus faster than 8Mhz?
    If you can change the speed (faster or slower), see if that
    makes a difference. Most NE2000 clones will run at 16Mhz, but
    some may not. Changing speed can also mask a noisy bus.

    What other devices are on the bus?
    If moving the devices around changes the reliability, then you
    have a bus noise problem -- just what that error message was
    designed to detect. Congratulations, you've probably found the
    source of other problems as well.

    Machine Hangs during Boot.
    ==========================

    Problem:  The machine hangs during boot right after the "8390..."  or
          "WD...." message. Removing the NE2000 fixes the problem.

    Solution: Change your NE2000 base address to 0x360 (or 0x340 for
          pl12 or later kernels.) Alternatively, you can use the new
          device registrar implemented in pl13 (see section 5.1)

    Reason:   Your NE2000 clone isn't a good enough clone. An active
          NE2000 is a bottomless pit that will trap any driver
          autoprobing in its space. The other ethercard drivers take
          great pain to reset the NE2000 so that it's safe, but some
          clones cannot be reset. Clone chips to watch out for:
          Winbond 83C901. Changing the NE2000 to a less-popular
          address will move it out of the way of other autoprobes,
          allowing your machine to boot.

    Problem:  The machine hangs during the SCSI probe at boot.

    Solution: It's the same problem as above, change the
          ethercard's address, or use the device registrar.

    Problem:  The machine hangs during the soundcard probe at boot.

    Solution: No, that's really during the silent SCSI probe, and it's
          the same problem as above.

    "eth0: DMAing conflict in ne_block_input"
    =========================================

    This bug came from timer-based packet retransmissions. If you got a
    timer tick _during_ a ethercard RX interrupt, and timer tick tried to
    retransmit a timed-out packet, you could get a conflict. Because of
    the design of the NE2000 you would have the machine hang (exactly the
    same the NE2000-clone boot hangs).

    Early versions of the driver disabled interrupts for a long time,
    and didn't have this problem. Later versions are fixed. (ie. kernels
    after 0.99p9 should be OK.)

    NE2000 not detected at boot.
    ============================

    A few people have reported a problem with detecting the Accton NE2000.
    This problem occurs only at boot-time, and the card is later detected
    at run-time by the identical code my (alpha-test) ne2k diagnostic
    program. Accton has been very responsive, but I still haven't tracked
    down what is going on. I've been unable to reproduce this problem
    with the Accton cards we purchased. If you are having this problem,
    please send me an immediate bug report. For that matter, if you have
    an Accton card send me a success report, including the type of the
    motherboard. I'm especially interested in finding out if this problem
    moves with the particular ethercard, or stays with the motherboard.

    Here are some things to try, as they have fixed it for some people:
    1) Change the bus speed, or just move the card to a different slot (!).
    2) Change the "I/O recovery time" parameter in the BIOS
       chipset configuration.
    3) Make the following code change suggested by David Cutler,
       <[e-mail]> to ne.c around line 150:

        for(i = 0; i < 32 /*sizeof(SA_prom)*/; i+=2) {
    -        SA_prom[i] = inb_p(ioaddr + NE_DATAPORT);
    -        SA_prom[i+1] = inb_p(ioaddr + NE_DATAPORT);
    +        SA_prom[i] = inb(ioaddr + NE_DATAPORT);
    +        SA_prom[i+1] = inb(ioaddr + NE_DATAPORT);
            if (SA_prom[i] != SA_prom[i+1])
                wordlength = 1;
        }

    Yes, this removes the delay between board accesses, something that
    would normally increase the likelihood of data corruption rather
    than decreasing it. Note that this change is already incorporated
    into pl15. If you have an older kernel, you may have to do it
    yourself.

6.02 Problems with WD80*3 cards

    Detected Non-existent Ethercard
    ===============================

    Problem:  A WD80*3 is falsely detected. Removing the sound or
          MIDI card eliminates the "detected" message.

    Solution: Update your ethercard driver: new versions include an
          additional sanity check.

    Reason:   Some MIDI ports happen to produce the same checksum as a
          WD ethercard.

    Error messages from the 80*3
    ============================

    Problem:  You get messages such as the following with your 80*3:
            eth0: bogus packet size, status = ........
            kmalloc called with impossibly large argument (65400)
            eth0: Couldn't allocate sk_buff of size 65400
            eth0: receiver overrun

    Reason:   There is a shared memory problem.

    Solution: If the problem is sporadic, you have hardware problems.
          Typical problems that are easy to fix are board conflicts,
          having cache or "shadow ROM" enabled for that region, or
          running your bus faster than 8Mhz. There are also a
          surprising number of memory failures on ethernet cards,
          so run a diagnostic program if you have one for your
          ethercard.

          If the problem is continual, and you have have to reboot
          to fix the problem, record the boot-time probe message
          and mail it to [e-mail] - Take particular note of
          the shared memory location.

    Will not detect my 80x3
    =======================

    Reason:   The Mitsumi CD-ROM (mcd) driver probe at 0x300 will
          succeed if just about *anything* is that I/O location.
          This is bad news and needs to be a bit more robust. (pl15)
          Once another driver registers that it "owns" an I/O
          location, other drivers (incl. the wd80x3) are "locked
          out" and can not probe that addr for a card.

    Solution: Recompile a new kernel without any excess drivers that
          you aren't using, including the above mcd driver.
          Or try moving your ethercard to a new I/O addr. Valid
          I/O addr. for all the cards are listed in section 5.1
          You can also point the mcd driver off in another direction
          by a boot-time parameter (via LILO) such as:
            "mcd=0x200,12"

6.03 Problems with 3Com cards

    Choosing the Interrupt of the 3c503
    ===================================

    Problem:  The 3c503 picks IRQ n at boot, but this is needed for some
          other device which needs IRQ n. (eg. CD ROM driver, etc.)
          Can this be fixed without compiling this into the kernel?

    Solution: The 3c503 driver probes for a free IRQ line in the order
          {5, 9/2, 3, 4}, and it should pick a line which isn't being
          used. The pre-pl12 (SLS 1.02) driver picked the IRQ line
          at boot-time, and the current driver (pl12) chooses when
          the card is open()/'ifconfig'ed. Note the "bug" noted in
          the 3c503 section in 1.01

          Alternately, you can fix the IRQ at boot by passing
          parameters via LILO. The following selects IRQ9, base
          location 0x300, <ignored value>, and if_port #1 (the
          external transceiver).
            lilo: linux ether=9,0x300,0,1,eth0

          The following selects IRQ3, probes for the base location,
          <ignored value>, and the default if_port #0 (the internal
          transceiver)
            lilo: linux ether=3,0,0,0,eth0

    "3c503: Configured interrupt number XX is out of range."
    ========================================================

    Problem:  Whoever built your kernel fixed the ethercard IRQ at XX.

    Reason:   The above is truly evil, and worse than that, it is
          not necessary. The 3c503 will autoIRQ when it gets
          "ifconfig"ed, and pick one of IRQ{5, 2/9, 3, 4}.

    Solution: Use lilo to set the IRQ, or rebuild the kernel, enabling
          autoIRQ by not specifying the IRQ line.

    Choosing the output of the 3c503
    ================================

    Problem:  The supplied 3c503 drivers don't use the AUI (thicknet) port.
          How does one choose it over the default thinnet port?

    Solution: The 3c503 AUI port can be selected at boot-time with 0.99pl12
          and later. The selection is overloaded onto the low bit of
          the currently-unused dev->rmem_start variable, so a boot-time
          parameter of:
            lilo: linux ether=0,0,0,1,eth0
          should work. A boot line to force IRQ 5, port base 0x300,
          and use an external transceiver is:
            lilo: linux ether=5,0x300,0,1,eth0

7 Networking with a laptop computer

    There are currently only a few ways to put your laptop on a network.
    You can use the SLIP code (and run at serial line speeds);
    you can buy one of the few laptops that come with a NE2000-compatible
    ethercard or PCMCIA slot built-in; you can get a laptop with a
    docking station and plug in an ISA ethercard; or you can use a
    parallel port Ethernet adapter such as the D-Link DE-600.

7.01 Option 1 -- using SLIP

    This is the cheapest solution, but by far the most difficult. Also,
    you will not get very high transmission rates. Since SLIP is not
    really related to ethernet cards, it will not be discussed further
    here. See the NET-2 HOWTO.

7.02 Option 2 -- Built in NE2000 compatible or PCMCIA Ethercard.

    The second solution severely limits your laptop choices and is fairly
    expensive. Be sure to read the specifications carefully, you may find
    that you will have to buy an additional non-standard transceiver to
    actually put the machine on a network.

    As this area of Linux development is fairly young, I'd suggest
    that you join the LAPTOPS mailing channel. See section 0.02
    which describes how to join a mailing list channel. Try and
    determine exactly what hardware you have (ie. card manufacturer,
    PCMCIA chip controller manufacturer) and then ask on the LAPTOPS
    channel. Regardless, don't expect things to be all that simple.
    Expect to have to fiddle around a bit, and patch kernels, etc.
    Maybe someday you will be able to type "make config" 8-)

    There is a number of programs on tsx-11.mit.edu in
    /pub/linux/packages/laptops/ that you may find useful. These
    range from PCMCIA Ethercard drivers to programs that communicate
    with the PCMCIA controller chip. Note that these drivers are
    usually tied to a specific PCMCIA chip (ie. the intel 82365
    or the TCIC/2) On a brighter note, I know that people have used the
    Linksys/D-Link 650 PCMCIA Ethernet PC Card with both controller
    chipsets, so it *can* be done. I have also seen reports of
    people using the IBM Credit Card Adapter for Ethernet with the
    intel 82365 chip. This is all just from following the LAPTOPS
    channel.

    The 3c589 from 3Com is supposed to look a lot like a 3c509.

    Anyway, the PCMCIA driver problem isn't specific to the Linux world.
    It's been a real disaster in the MS-DOS world. In that world
    people expect the hardware to work if they just follow the manual.
    They might not expect it to interoperate with any other hardware
    or software, or operate optimally, but they do expect that the
    software shipped with the product will function. Many PCMCIA
    adaptors don't pass this test.

7.03 Option 3 -- ISA Ethercard in the Docking Station.

    I recommend the third solution. Docking stations for laptops typically
    cost about $250 and provide two full-size ISA slots, two serial and one
    parallel port. Most (all?) docking stations are powered off of the
    laptop's batteries, and a few allow adding extra batteries in the
    docking station if you use short ISA cards. You can add an inexpensive
    ethercard and enjoy full-speed ethernet performance.

7.04 Option 4 -- Pocket / parallel port adaptors.

    The "pocket" ethernet adaptors may also fit your need.
    Until recently they actually costed more than a docking station and
    cheap ethercard, and most tie you down with a wall-brick power supply.
    At present, you can choose from the D-Link, or the RealTek adaptor.
    Most other companies, especially Xircom, treat the programming
    information as a trade secret, so support will likely be slow in
    coming.

    You can sometimes avoid the wall-brick with the adaptors by buying
    or making a cable that draws power from the laptop's keyboard
    port. (This is mentioned in the info. for the AT-Lan-Tec unit.)

    The keyboard pinouts (5 pin DIN) are as follows:

        Signal/Function            Pin #
        ---------------            -----
        KEYCLK (clock)            1
        KEYDAT (data)            2
        N/C                3
        Ground                4
        +5 V                5

    A quick check with a voltmeter will verify which pins are 4 and 5
    if you are not sure.
8 Frequently asked questions

    Here are some of the more frequently asked questions about using
    Linux with an Ethernet connection. Some of the more specific
    questions are sorted on a "per manufacturer basis" and are listed
    in the "Troubleshooting" section. (section 6). However, since this
    document is basically "old" by the time you get it, any "new" problems
    will not appear here instantly. For these, I suggest that you make
    efficient use of your newsreader. For example, nn users would type
        nn -xX -s'3c'
    to get all the news articles in your subscribed list that have
    "3c" in the subject. (ie. 3com, 3c509, 3c503, etc.)
    The moral: Read the man page for your newsreader.

8.01 Just the FAQ's ma'am -- just the FAQ's.

    Q: I heard that there is an alpha driver available for my card.
       Where can I get it?

    A: Assuming that it is a djb driver, it will be on ftp.super.org
       in the /pub/linux/ area. Things change here quite frequently,
       so just look around for it. There is usually only about 3
       subdirs, so you should be able to find it. Now, if it really
       is an alpha, or pre-alpha driver, then please treat it as
       such. In other words, don't complain because you can't figure
       out what to do with it. If you can't figure out how to install
       it, then you probably shouldn't be testing it. Also, if it brings
       your machine down, don't complain. Instead, send us a well
       documented bug report, or even better, a patch!

    Q: Is there token ring support for Linux?

    A: No, there is no token ring support in Linux. To support token ring
       requires more than only a writing a device driver, it also requires
       writing the source routing routines for token ring. Given that
       token ring is expensive, not fast, and will probably be swept away
       by 100baseVG in a few months, it doesn't seem worth it to write
       a driver. In case anyone wants to, I looked at writing a token ring
       device driver, and concluded that the hardware interface
       wasn't too difficult to do, but writing the support for source
       routing would take significantly longer than I was willing to spend
       on an expensive and dying technology.

       Alan Cox adds: "It will require [...] changes to the bottom socket
       layer to support 802.2 and 802.2 based TCP/IP. Don't expect
       anything soon."

    Q: Is there FDDI support for Linux?

    A: No, there is no Linux driver for any FDDI boards. I come from a
       place with supercomputers, so an external observer might think
       FDDI would be high on my list. But FDDI never delivered end-to-end
       throughput that would justify its cost, and it seems to be a nearly
       abandoned technology now that 100base{X,Anynet} seems imminent.
       (And yes, I know you can now get FDDI boards for <$1K. That
       seems to be a last-ditch effort to get some return on the
       development investment. Where is the next generation of FDDI
       going to come from?)

    Q: Can I link 10BaseT (RJ45) based systems together without a hub?

    A: You can link 2 machines easily, but no more than that, without
       extra devices/gizmos. See section 4 on wiring -- it explains
       how to do it. And no, you can't hack together a hub just by
       crossing a few wires and stuff. It's pretty much impossible
       to do the collision signal right without duplicating a hub.

    Q: Is there IPX or Novell support available for Linux?

    A: Alan Cox writes: "The novell protocols are available from novell
       for various amounts. IPX is freely documented. SPX is about $1000
       but I'm told Xerox SPP is identical. _PLEASE_ has anyone got any
       freely distributable Xerox SPP code/documentation? The novell
       server spec costs you $15000 + royalties providing you only
       want to write a client, or $30000 + royalties otherwise. Needless
       to say the final output has to be binary only and subject to a
       novell license. Reading their license rules by my interpretation
       its also impossible for us to do because you would seem to have
       to bar disassembly of your final result, which is not allowed
       in the EEC.

       Bits of NCP are known, and I hope eventually enough will be known
       to write limited NCP support into Linux, for the moment I'm poking
       around at IPX, tho this will have to wait until the new network
       code is finished.

       An Alpha test IPX protocol layer is available from me (Alan)
       for pl14 or higher. People are also exploring the issue of NCP and
       the new Dr Dobbs journal article on the innards of netware has
       provided a core of good information."

       As an alternative, Miquel van Smoorenburg suggests the following:
       "It _is_ possible to set up a dedicated PC running both novell and
       the PD SOSS server and let it gateway from NFS to novell. This way
       it is possible to mount the Novell drives on the Unix client.

       SOSS is a PD (perhaps with some restrictions, but freely available)
       NFS server for DOS. It includes the PC/IP TCP/IP implementation
       and runs on a packet driver. I have run both a Novell client
       (with PDIPX, a Packet Driver IPX) and this SOSS server together
       successfully."

       You can get "Stan's Own Server System" from the following location:
        hilbert.wharton.upenn.edu:pub/tcpip/soss.zip
       Note that this version has the IP bugs fixed and the subdirs
       with extensions bug fixed. Some of the soss.zoo archives do
       not contain these fixes.

    Q: What needs to be done so that Linux can run two ethernet cards?

    A: The easiest solution is to get 0.99pl13 or newer, as the hooks for
       multiple ethercards are all there.
       You can enable additional ethercards with LILO parameters such as:

        lilo: linux ether=5,0x300,0,1,eth0 ether=15,0x280,eth1

       These boot time arguments can be made permanent so that you
       don't have to re-enter them every time. See the LILO manual.

    Q: Okay, I can run 2 cards -- can I run Linux as a gateway
       between two networks?

    A: This is really a question for the NET-HOWTO, but it is
       answered here anyways: Charles Hedrick (aka Mr. Slip)
       had this to say:

       "Yes, however I'm a bit nervous about doing it. The problem isn't
       functionality -- there's IP forwarding code, and as far as I know,
       it works. Some people do use it. However routers need to be
       particularly careful to avoid creating network problems such as
       "meltdowns." The Linux IP layer doesn't have quite enough of these
       protective features. It will only cause trouble if other hosts on
       your network are misconfigured, and even then it probably won't
       cause much trouble (assuming that only systems actually acting as
       gateways are built with IP_FORWARD enabled). But I'd still rather
       use a router that met all of the requirements of the host and
       router requirements in the RFC's. (Note that not all other Unix
       implementations do either. I'm concerned about things like not
       sending ICMP responses to messages that arrive as media
       broadcasts. 386BSD looks OK, but older BSD-based implementations
       often didn't do all of these checks.)

       It depends a lot on what the network is like and how critical it is.
       For a home setup with a couple of hosts, I see no problem at all.
       But I would not consider using Linux as a router on a large
       campus network at the moment. I still think that by release 1.0,
       Linux will be a reasonably well-behaved host. But I think use
       as a router in critical situations should wait until somebody
       has checked the ip and icmp modules for compliance with RFC 1009
       and a few other specs."

    Q: I have /dev/eth0 as a link to /dev/xxx. Is this right?

    A: Contrary to what you have heard, the files in /dev/* are not used.
       I originally thought that they might be an OK idea. I've since
       concluded that they won't work, at least in the documented form.

    Q: Should I disable trailers when I "ifconfig" my ethercard?

    A: You can't disable trailers, and you shouldn't want to.
       'Trailers' are a hack to avoid data copying in the
       networking layers. The idea was to use a trivial
       fixed-size header of size 'H', put the variable-size header
       info at the end of the packet, and allocate all packets
       'H' bytes before the start of a page. While it was a good
       idea, it turned out to not work well in practice.
       If someone suggests the use of '-trailers', note that it
       is the equivalent of sacrificial goats blood. It won't do
       anything to solve the problem, but if problem fixes itself then
       someone can claim deep magical knowledge.

9 Miscellaneous.

    Any other associated stuff that didn't fit in anywhere else gets
    dumped here. It may not be relevant, and it may not be of general
    interest but it is here anyway.

9.01 Bad Vendors

    #define SOAPBOX

    There used to be some horror stories here about dealings with
    Cabletron and Xircom. They were pretty ugly and gruesome.
    Basically these companies are the ethernet equivalent of
    what Diamond is to XFree86. They do not want to release
    vital information on low-level programming of their hardware.
    For something like Linux, where the source code for everything
    is out in the open, this makes their hardware difficult or
    impossible to use. However, like Diamond, when confronted
    with the fact that they are losing sales from Linux/BSD users,
    they basically shrug it off, saying that it is only a small
    percentage of the total sales. If you can afford the time,
    drop these vendors a note (via e-mail or snail-mail) and tell
    them politely that the fact that they don't support open
    software systems such as Linux has forced you to exclude them
    from the vendors that you are purchasing hardware from. It may
    not make any immediate difference, but it might make you feel
    better. Besides, a few seconds of your time is a cheap price
    to pay for *all* that free Linux software you are using. 8-)

    #undef SOAPBOX

9.02 Closing

    If you have found any glaring typos, or outdated info in this
    document, please let one of us know. It's getting big, and it
    is easy to overlook stuff.

    Paul Gortmaker        <[e-mail]>
    Donald J. Becker    <[e-mail]>

        =========== end of Ethernet HOWTO ============