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'. Chances are the question you want an answer for has already been asked (and answered!) by someone else, so even if you don't find your answer here, you probably can find what you want from a news archive such as Dejanews.
With most Linux distributions, the drivers exist as loadable
modules, which are small binary files that are merged with
the operating system at run time. A module gives the
operating system (kernel) the information on how to control
that particular ethernet card. The name of each module is
listed in the heading of the section for each card in this
document. Once you know the name of the module, you have to
add it to the file /etc/modules.conf
so Linux will
know what module to load for your card. The syntax is typically
as follows.
alias eth0 module_name options module_name option1=value1 option2=value2 ...
The options line is typically only needed for older ISA hardware.
For multiple card systems, additional lines with eth1
,
eth2
and so on are usually required.
The module files typically live in the directory
/lib/modules/
which is further subdivided by kernel
version (use uname -r
) and subsystem (in this case
net
). These are put there by the distribution
builder, or by the individual user when they run
make modules_install
after building their own
kernel and modules (see the kernel howto for more details
on building your own stuff).
If you build your own kernel, you have the option of having
all the drivers merged with the kernel right then and there,
rather than existing as separate files. When this is done,
the drivers will detect the hardware at boot up. Options
to the drivers are supplied by the kernel command line prior
to boot (see BootPrompt Howto for more details). The user
chooses what drivers are used during the make config
step of building the kernel (again see the kernel howto).
The answer to this question depends heavily on exactly what you intend on doing with your net connection, and how much traffic it will see.
If you only expect a single user to be doing the occasional ftp session or WWW connection, then even an old ISA card will probably keep you happy (assuming 10Mbps, not 100).
If you intend to set up a server, and you require the CPU overhead of moving data over the network to be kept to a minimum, you probably want to look at one of the PCI cards that uses a chip with bus-mastering capapbility. In addition, some cards now can actually do some of the processing overhead of data checksums right on the card, giving the CPU even more of a break. For more details please see:
Hardware Checksum/Zerocopy Page
If you fall somewhere in the middle of the above, then any one of the low cost PCI cards with a stable driver will do the job for you.
I heard that there is an updated or preliminary/alpha driver available for my card. Where can I get it?
The newest of the `new' drivers can be found on Donald's
WWW site: www.scyld.com
- things
change here quite frequently, so just look around for it.
Alternatively, it may be easier to use a WWW browser on:
to locate the driver that you are looking for. (Watch out for WWW browsers that silently munge the source by replacing TABs with spaces and so on - use ftp, or at least an FTP URL for downloading if unsure.)
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!
Note that some of the `useable' experimental/alpha drivers have
been included in the standard kernel source tree. When running
make config
one of the first things you will be asked
is whether to ``Prompt for development and/or incomplete
code/drivers''. You will have to answer `Y' here to get
asked about including any alpha/experiemntal drivers.
What needs to be done so that Linux can run two or more ethernet cards?
The answer to this question depends on whether the driver(s)
is/are being used as a loadable module or are compiled directly
into the kernel. Most linux distributions use modular drivers now.
This saves distributing lots of kernels, each with a different driver
set built in. Instead a single basic kernel is used and the
individual drivers that are need for a particular user's system are
loaded once the system has booted far enough to access
the driver module files (usually stored in /lib/modules/
).
In the case of PCI cards, the PCI drivers/modules should detect all of the installed cards that it supports automatically. The user should not supply any I/O base or IRQ information, unless specifically instructed to do so by the individual driver documentation in order to support some non-standard machine.
Some earlier kernels had a limit of 16 ethercards that could be detected at boot, and some ISA modules have a limit of four cards per loaded module. You can always load another copy of the same module under a different name to support another four cards if this is a limitation, or recompile the module with support for as many as you require.
For ISA cards, probing for a card is not a safe operation, and
hence you typically need to supply the I/O base address of the
card so the module knows where to look. This information is
stored in the file /etc/modules.conf
.
As an example, consider a user that has two ISA NE2000 cards,
one at 0x300
and one at 0x240
and what lines they
would have in their /etc/modules.conf
file:
alias eth0 ne alias eth1 ne options ne io=0x240,0x300
What this does: This says that if the administrator (or the
kernel) does a modprobe eth0
or a modprobe eth1
then
the ne.o
driver should be loaded for either eth0
or
eth1
. Furthermore, when the ne.o
module is loaded, it
should be loaded with the options io=0x240,0x300
so that the
driver knows where to look for the cards. Note that the 0x
is important - things like 300h
as commonly used in the DOS
world won't work. Switching the order of the 0x240
and
the 0x300
will switch which physical card ends up as
eth0
and eth1
.
Most of the ISA module drivers can take multiple comma separated
I/O values like this example to handle multiple cards. However,
some (older?) drivers, such as the 3c501.o module are currently
only able to handle
one card per module load. In this case you can load the module
twice to get both cards detected. The /etc/modules.conf
file in this case would look like:
alias eth0 3c501 alias eth1 3c501 options eth0 -o 3c501-0 io=0x280 irq=5 options eth1 -o 3c501-1 io=0x300 irq=7
In this example the -o
option has been used to give each
instance of the module a unique name, since you can't have two
modules loaded with the same name. The irq=
option has
also been used to to specify the hardware IRQ setting of the card.
(This method can also be used with modules that accept comma
separated I/O values, but it is less efficient since the module
ends up being loaded twice when it doesn't really need to be.)
As a final example, consider a user with one 3c503 card
at 0x350
and one SMC Elite16 (wd8013) card at 0x280
.
They would have:
alias eth0 wd alias eth1 3c503 options wd io=0x280 options 3c503 io=0x350
For PCI cards, you typically only need the alias
lines to
correlate the ethN
interfaces with the appropriate driver
name, since the I/O base of a PCI card can be safely detected.
The available modules are typically stored in
/lib/modules/`uname -r`/net
where the
uname -r
command gives the kernel version (e.g. 2.0.34).
You can look in there to see which one matches your card.
Once you have the correct settings in your modules.conf
file, you can test things out with:
modprobe eth0 modprobe eth1 ... modprobe ethN-1
where `N' is the number of ethernet interfaces you have.
Note that the interface name (ethX
) assigned to the driver
by the kernel is independent of the name used on the alias line.
For further details on this, see:
Using the Ethernet Drivers as Modules
Since some ISA card probes can hang the machine, kernels up to and including 2.4 only autoprobe for one ISA ethercard by default. As there aren't any distribution kernels with lots of ISA drivers built in anymore, this restriction is no longer in 2.6 and newer.
As of 2.2 and newer kernels, the boot probes have been sorted into safe and unsafe, so that all safe (e.g. PCI and EISA) probes will find all related cards automatically. Systems with more than one ethernet card with at least one of them being an ISA card will still need to do one of the following.)
There are two ways that you can enable auto-probing for
the second (and third, and...) card. The easiest
method is to pass boot-time arguments to the kernel,
which is usually done by LILO. Probing for the
second card can be achieved by using a boot-time argument
as simple as ether=0,0,eth1
. In this
case eth0
and eth1
will be assigned in the order
that the cards are found at boot. Say if you want
the card at 0x300
to be eth0
and
the card at 0x280
to be eth1
then you could use
LILO: linux ether=5,0x300,eth0 ether=15,0x280,eth1
The ether=
command accepts more than the IRQ + I/O
+ name shown above. Please have a look at
Passing Ethernet Arguments...
for the full syntax, card specific parameters, and LILO tips.
The second way (not recommended) is to edit the file
Space.c
and replace the 0xffe0
entry for the
I/O address with a zero. The 0xffe0
entry tells it
not to probe for that device -- replacing it with a zero
will enable autoprobing for that device.
ether=
thing didn't do anything for me. Why?
As described above, the ether=
command only works
for drivers that are compiled into the kernel. Now most
distributions use the drivers in a modular form, and so
the ether=
command is rarely used anymore. (Some older
documentation has yet to be updated to reflect this change.)
If you want to apply options for a modular ethernet driver
you must make changes to the /etc/modules.conf
file.
If you are using a compiled in driver, and have added
an ether=
to your LILO configuration file, note
that it won't take effect until you re-run lilo
to process the updated configuration file.
Problem: PCI NE2000 clone card is not detected at boot with v2.0.x.
Reason:
The ne.c
driver up to v2.0.30 only knows about the PCI
ID number of RealTek 8029 based clone cards. Since then,
several others have also released PCI NE2000 clone
cards, with different PCI ID numbers, and hence the
driver doesn't detect them.
Solution: The easiest solution is to upgrade to a v2.0.31 (or newer) version of the linux kernel. It knows the ID numbers of about five different NE2000-PCI chips, and will detect them automatically at boot or at module loading time. If you upgrade to 2.0.34 (or newer) there is a PCI-only specific NE2000 driver that is slightly smaller and more efficient than the original ISA/PCI driver.
Problem: PCI NE2000 clone card is reported as an ne1000 (8 bit card!) at boot or when I load the ne.o module for v2.0.x, and hence doesn't work.
Reason: Some PCI clones don't implement byte wide access (and hence are not truly 100% NE2000 compatible). This causes the probe to think they are NE1000 cards.
Solution: You need to upgrade to v2.0.31 (or newer) as described above. The driver(s) now check for this hardware bug.
Problem: PCI NE2000 card gets terrible performance, even when reducing the window size as described in the Performance Tips section.
Reason: The spec sheets for the original 8390 chip, desgined and sold over ten years ago, noted that a dummy read from the chip was required before each write operation for maximum reliablity. The driver has the facility to do this but it has been disabled by default since the v1.2 kernel days. One user has reported that re-enabling this `mis-feature' helped their performance with a cheap PCI NE2000 clone card.
Solution:
Since it has only been reported as a solution by one person, don't
get your hopes up. Re-enabling the read before write fix is done
by simply editing the driver file in linux/drivers/net/
,
uncommenting the line containing NE_RW_BUGFIX
and then
rebuilding the kernel or module as appropriate. Please send an
e-mail describing the performance difference and type of card/chip
you have if this helps you. (The same can be done for the
ne2k-pci.c
driver as well).
Problem:
The ne2k-pci.c
driver reports error messages like
timeout waiting for Tx RDC
with a PCI NE2000 card and doesn't
work right.
Reason: Your card and/or the card to PCI bus link can't handle the long word I/O optimization used in this driver.
Solution:
Firstly, check the settings available in the BIOS/CMOS setup
to see if any related to PCI bus timing are too aggressive for
reliable operation. Otherwise using the ISA/PCI ne.c
driver (or removing the #define USE_LONGIO
from
ne2k-pci.c
) should let you use the card.
Probem: ISA Plug and Play NE2000 (such as RealTek 8019) is not detected.
Reason: The original NE2000 specification (and hence the linux NE2000 driver in older kernels) did not have support for Plug and Play.
Solution:
Either use a 2.4 or newer kernel that has a NE2000 driver with PnP,
or use the DOS configuration disk that came with the card to disable
PnP, and to set the card to a specified I/O address and IRQ. Add
a line to /etc/modules.conf
like options ne io=0xNNN
where 0xNNN
is the hex I/O address you set the card to. (This
assumes you are using a modular driver; if not then use an
ether=0,0xNNN,eth0
argument at boot). You may also have to
enter the BIOS/CMOS setup and mark the IRQ as Legacy-ISA instead of
PnP.
Problem: NE*000 driver reports `not found (no reset ack)' during boot probe.
Reason: This is related to the above change. After the initial verification that an 8390 is at the probed I/O address, the reset is performed. When the card has completed the reset, it is supposed to acknowedge that the reset has completed. Your card doesn't, and so the driver assumes that no NE card is present.
Solution:
You can tell the driver that you have a bad card by using
an otherwise unused mem_end
hexidecimal value of 0xbad
at
boot time. You have to also supply a non-zero I/O base
for the card when using the 0xbad
override. For example,
a card that is at 0x340
that doesn't ack the reset
would use something like:
LILO: linux ether=0,0x340,0,0xbad,eth0
This will allow the card detection to continue, even if your
card doesn't ACK the reset. If you are using the driver as
a module, then you can supply the option bad=0xbad
just
like you supply the I/O address.
Problem: NE*000 card hangs machine at first network access.
Reason: This problem has been reported for kernels as old as 1.1.57 to the present. It appears confined to a few software configurable clone cards. It appears that they expect to be initialized in some special way.
Solution: Several people have reported that running the supplied DOS software config program and/or the supplied DOS driver prior to warm booting (i.e. loadlin or the `three-finger-salute') into linux allowed the card to work. This would indicate that these cards need to be initialized in a particular fashion, slightly different than what the present Linux driver does.
Problem:
NE*000 ethercard at 0x360
doesn't get detected.
Reason:
Your NE2000 card is 0x20
wide in
I/O space, which makes it hit the parallel port at 0x378
.
Other devices that could be there are the second floppy
controller (if equipped) at 0x370
and the secondary
IDE controller at 0x376--0x377
.
If the port(s) are already registered by another driver, the
kernel will not let the probe happen.
Solution:
Either move your card to an address like 0x280, 0x340, 0x320
or compile without parallel printer support.
Problem: Network `goes away' every time I print something (NE2000)
Reason: Same problem as above, but you have an older kernel that doesn't check for overlapping I/O regions. Use the same fix as above, and get a new kernel while you are at it.
Problem: NE*000 ethercard probe at 0xNNN: 00 00 C5 ... not found. (invalid signature yy zz)
Reason:
First off, do you have a NE1000 or NE2000 card at the addr. 0xNNN?
And if so, does the hardware address reported look like a valid
one? If so, then you have a poor NE*000 clone. All NE*000 clones
are supposed to have the value 0x57
in bytes 14 and 15 of the
SA PROM on the card. Yours doesn't -- it has `yy zz' instead.
Solution:
There are two ways to get around this. The easiest is to
use an 0xbad
mem_end value as described above for the
`no reset ack' problem. This will bypass the signature check,
as long as a non-zero I/O base is also given. This way no
recompilation of the kernel is required.
The second method (for hackers) involves changing the driver
itself, and then recompiling your kernel (or module).
The driver (/usr/src/linux/drivers/net/ne.c) has a "Hall of Shame"
list at about line 42. This list is used to detect poor clones.
For example, the DFI cards use `DFI' in the first 3 bytes of the
PROM, instead of using 0x57
in bytes 14 and 15, like they are
supposed to.
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 something like 0x340
.
Alternatively, you
can use the ``reserve='' boot argument in conjunction with
the ``ether='' argument to protect the card from other
device driver probes.
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. 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.
Reason: It's the same problem as above, change the ethercard's address, or use the reserve/ether boot arguments.
Problem: The machine hangs during the soundcard probe at boot.
Reason: No, that's really during the silent SCSI probe, and it's the same problem as above.
Problem: NE2000 not detected at boot - no boot messages at all
Solution: There is no `magic solution' as there can be a number of reasons why it wasn't detected. The following list should help you walk through the possible problems.
1) Build a new kernel with only the device drivers that you need.
Verify that you are indeed booting the fresh kernel. Forgetting to
run lilo, etc. can result in booting the old one. (Look closely at
the build time/date reported at boot.) Sounds obvious, but we have
all done it before. Make sure the driver is in fact included in
the new kernel, by checking the System.map
file for names
like ne_probe
.
2) Look at the boot messages carefully. Does it ever even mention
doing a ne2k probe such
as `NE*000 probe at 0xNNN: not found (blah blah)'
or does it just fail silently. There is a big difference.
Use dmesg|more
to review the boot messages after logging in, or hit Shift-PgUp
to scroll the screen up after the boot has completed and the login
prompt appears.
3) After booting, do a cat /proc/ioports
and verify
that the full iospace that the card will require is vacant. If
you are at 0x300
then the ne2k driver will ask
for 0x300-0x31f
. If any other device driver has registered
even one port anywhere in that range, the probe will not
take place at that address and will silently continue to the next
of the probed addresses. A common case is having the lp driver
reserve 0x378
or the second IDE channel reserve 0x376
which stops the ne driver from probing 0x360-0x380
.
4) Same as above for cat /proc/interrupts
. Make sure no
other device has registered the interrupt that you set
the ethercard for. In this case, the probe will happen, and the
ether driver will complain loudly at boot about not being able to
get the desired IRQ line.
5) If you are still stumped by the silent failure of the driver, then
edit it and add some printk() to the probe. For example, with the ne2k
you could add/remove lines (marked with a `+' or `-') in
linux/drivers/net/ne.c
like:
int reg0 = inb_p(ioaddr); + printk("NE2k probe - now checking %x\n",ioaddr); - if (reg0 == 0xFF) + if (reg0 == 0xFF) { + printk("NE2k probe - got 0xFF (vacant I/O port)\n"); return ENODEV; + }
Then it will output messages for each port address that it checks, and you will see if your card's address is being probed or not.
6) You can also get the ne2k diagnostic from Don's ftp site (mentioned
in the howto as well) and see if it is able to detect your card after
you have booted into linux. Use the `-p 0xNNN
' option to tell it
where to look for the card. (The default is 0x300
and it doesn't
go looking elsewhere, unlike the boot-time probe.)
The output from when it finds a card will look something like:
Checking the ethercard at 0x300. Register 0x0d (0x30d) is 00 Passed initial NE2000 probe, value 00. 8390 registers: 0a 00 00 00 63 00 00 00 01 00 30 01 00 00 00 00 SA PROM 0: 00 00 00 00 c0 c0 b0 b0 05 05 65 65 05 05 20 20 SA PROM 0x10: 00 00 07 07 0d 0d 01 01 14 14 02 02 57 57 57 57 NE2000 found at 0x300, using start page 0x40 and end page 0x80.
Your register values and PROM values will probably be different.
Note that all the PROM values are doubled for a 16 bit card, and
that the ethernet address (00:00:c0:b0:05:65) appears in the
first row, and the double 0x57
signature appears at the
end of the PROM.
The output from when there is no card installed at 0x300
will look something like this:
Checking the ethercard at 0x300. Register 0x0d (0x30d) is ff Failed initial NE2000 probe, value ff. 8390 registers: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff SA PROM 0: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff SA PROM 0x10: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff Invalid signature found, wordlength 2.
The 0xff
values arise because that is the value that
is returned when one reads a vacant I/O port. If you happen
to have some other hardware in the region that is probed, you
may see some non 0xff
values as well.
7) Try warm booting into linux from a DOS boot floppy (via loadlin) after running the supplied DOS driver or config program. It may be doing some extra (i.e. non-standard) "magic" to initialize the card.
8) Try Russ Nelson's ne2000.com packet driver to see if even it can see your card -- if not, then things do not look good. Example:
A:> ne2000 0x60 10 0x300
The arguments are software interrupt vector, hardware IRQ, and I/O base. You can get it from any msdos archive in pktdrv11.zip -- The current version may be newer than 11.
Problem: You get messages such as the following:
eth0: bogus packet size: 65531, status=0xff, nxpg=0xff
Reason: There is a shared memory problem.
Solution:
The most common reason for this is PCI machines that are
not configured to map in ISA memory devices. Hence you
end up reading the PC's RAM (all 0xff
values) instead of
the RAM on the card that contains the data from the
received packet.
Other typical problems that are easy to fix are board conflicts, having cache or `shadow ROM' enabled for that region, or running your ISA 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.
Problem: SMC EtherEZ doesn't work in non-shared memory (PIO) mode.
Reason: Older versions of the Ultra driver only supported the card in the shared memory mode of operation.
Solution: The driver in kernel version 2.0 and above also supports the programmed I/O mode of operation. Upgrade to v2.0 or newer.
Problem: Old wd8003 and/or jumper-settable wd8013 always get the IRQ wrong.
Reason: The old wd8003 cards and jumper-settable wd8013 clones don't have the EEPROM that the driver can read the IRQ setting from. If the driver can't read the IRQ, then it tries to auto-IRQ to find out what it is. And if auto-IRQ returns zero, then the driver just assigns IRQ 5 for an 8 bit card or IRQ 10 for a 16 bit card.
Solution: Avoid the auto-IRQ code, and tell the kernel what the IRQ that you have jumpered the card to in your module configuration file (or via a boot time argument for in-kernel drivers).
Problem: SMC Ultra card is detected as wd8013, but the IRQ and shared memory base is wrong.
Reason: The Ultra card looks a lot like a wd8013, and if the Ultra driver is not present in the kernel, the wd driver may mistake the ultra as a wd8013. The ultra probe comes before the wd probe, so this usually shouldn't happen. The ultra stores the IRQ and mem base in the EEPROM differently than a wd8013, hence the bogus values reported.
Solution: Recompile with only the drivers you need in the kernel. If you have a mix of wd and ultra cards in one machine, and are using modules, then load the ultra module first.
Problem: The 3c503 picks IRQ N, but this is needed for some other device which needs IRQ N. (eg. CD ROM driver, modem, 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 driver chooses when
the card is ifconfig
'ed into operation.
If you are using a modular driver, you can use module parameters to set various things, including the IRQ value.
The following selects IRQ9, base
location 0x300
, <ignored value>, and if_port #1 (the
external transceiver).
io=0x300 irq=9 xcvr=1
Alternately, if the driver is compiled into the kernel, you can set the same values at boot by passing parameters via LILO.
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
Problem: 3c503: configured interrupt X invalid, will use autoIRQ.
Reason: The 3c503 card can only use one of IRQ{5, 2/9, 3, 4} (These are the only lines that are connected to the card.) If you pass in an IRQ value that is not in the above set, you will get the above message. Usually, specifying an interrupt value for the 3c503 is not necessary. The 3c503 will autoIRQ when it gets ifconfig'ed, and pick one of IRQ{5, 2/9, 3, 4}.
Solution: Use one of the valid IRQs listed above, or enable autoIRQ by not specifying the IRQ line at all.
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 for in-kernel drivers, and at module insertion for modular drivers. 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 for in-kernel drivers.
To specify the AUI port when loading as a module, just append
xcvr=1
to the module options line along with
your I/O and IRQ values.
For best results (and minimum aggravation) it is recommended
that you use the (usually DOS) program that came with your
card to disable the ISA-PnP mechanism and set it to a fixed
I/O address and IRQ. Make sure the I/O address you use is
probed by the driver at boot, or if using modules then supply
the address as an io=
option in /etc/modules.conf
.
You may also have to enter the BIOS/CMOS setup and mark the IRQ
as Legacy-ISA instead of ISA-PnP (if your computer has this option).
Note that you typically don't need DOS installed to run a DOS based configuration program. You can usually just boot a DOS floppy disk and run them from the supplied floppy disk. You can also download OpenDOS and FreeDOS for free.
If you require ISA-PnP enabled for compatibility with some other
operating system then what you will have to do depends
on what kernel version you are using. For 2.2.x and older
kernels, you will have to use the isapnptools
package with linux to configure the card(s) each time at boot.
You will still have to make sure the I/O address chosen for the
card is probed by the driver or supplied as an io=
option.
For 2.4.x and newer kernels, there is ISA-PnP support available
built right into the kernel (if selected at build time) and
if your particular driver makes use of this support, then your
card will be configured to an available I/O address and IRQ
all without any user supplied option values. You do not
want to be trying to use the user-space isapnptools and the
in kernel ISA-PnP support at the same time.
Some systems have an `enable PnP OS' (or similar named) option in the BIOS/CMOS setup menu which does not really have anything to do with ISA-PnP hardware. See below for more details on this option.
Some PCI BIOSes may not enable all PCI cards at power-up, especially if the BIOS option `PnP OS' is enabled. This mis-feature is to support the current release of Windows which still uses some real-mode drivers. Either disable this option, or try and upgrade to a newer driver which has the code to enable a disabled card.
Note that kernel version 2.4.x has better support to deal with this option - in particular you should be able to enable this option, and the kernel/drivers should be able to set up and/or enable the cards on its own.
Version 1 of the PCI spec allowed for some slots to be bus master and some slots to be slave (non-bus master) slots. To avoid the problems associated with people putting BM cards into slave slots, the v2 of the PCI spec said that all slots should be BM capable. Hovever most PCI chipsets only have four BM pins, and so if you have a five slot board, chances are that two slots share one of the BM pins! This allows the board to meet the requirements of the v2 spec (but not the intent). So if you have a bunch of cards, and two of them fail to work, they may be in slots that share a BM pin.
/etc/conf.modules
and not /etc/modules.conf
.
Older distributions will have conf.modules
and not
modules.conf
which is the more sensible name of
the two. Newer module utility programs expect the new name,
so keep that in mind if you upgrade an older system.
The usual reason for this is that people are using a kernel that does not have support for their particular card built in. For a modular kernel, it usually means that the required module has not been requested for loading.
If you are using a modular based kernel, such as those installed
by most of the linux distributions, then try and use the
configuration utility for the distribution to select the module
for your card. For ISA cards, it is a good idea to determine
the I/O address of the card and add it as an
option (e.g. io=0x340
) if the configuration utility asks
for any options. If there is no configuration utility, then
you will have to add the correct module name (and options)
to /etc/modules.conf
-- see man modprobe
for
more details.
The next main cause is having another device using part
of the I/O space that your card needs. Most cards are
16 or 32 bytes wide in I/O space. If your card is set
at 0x300
and 32 bytes wide, then the driver will ask
for 0x300-0x31f
. If any other device driver has registered
even one port anywhere in that range, the probe will not
take place at that address and the driver will silently
continue to the next of the probed addresses. So, after
booting, do a cat /proc/ioports
and verify that the
full I/O space that the card will require is vacant.
Another problem is having your card jumpered to an I/O
address that isn't probed by default. The list of
probed addresses for each driver is easily found just
after the text comments in the driver source.
Even if the I/O setting of your card is
not in the list of probed addresses, you can supply
it at boot (for in-kernel drivers) with
the ether=
command as described in
Passing Ethernet Arguments...
Modular drivers can make use of the io=
option
in /etc/modules.conf
to
specify an address that isn't probed by default.
unresolved symbol ei_open
and won't load.
You are using one of the many ethernet cards that have
an 8390 chip (or clone) on board. For such cards, the
driver comes in two parts - the part which you unsuccessfully
tried to load (e.g. ne2k-pci.o, ne.o, wd.o, smc-ultra.o
etc.) and the 8390 part. These drivers have (+8390)
listed right beside their module name in the vendor
specific information listing.
(
Vendor Specific...)
You have to make sure that the
8390.o
module gets loaded before loading the
second half of the driver so that the second half of the
driver can find the functions in 8390.o
that it needs.
Possible causes: (1)Forgetting to run depmod after installing
a new kernel and modules, so that module dependencies like
this are handled automatically. (2)Using insmod
instead
of modprobe
, as insmod doesn't check for any module
ordering constraints. (3)The module 8390.o
is not
in the directory beside the other half of the driver where
it should be.
ifconfig
reports the wrong I/O address for the card.
No it doesn't. You are just interpreting it incorrectly.
This is not a bug, and the numbers reported are correct. It just
happens that some 8390 based cards (wd80x3, smc-ultra, etc) have the
actual 8390 chip living at an offset from the first assigned I/O port.
This is the value stored in
dev->base_addr
, and is what ifconfig
reports. If you
want to see the full range of ports that your card uses, then try
cat /proc/ioports
which will give the numbers you expect.
0xffff
)
This will usually show up as reads of lots of 0xffff
values.
No shared memory cards of any type will work in a PCI machine
unless you have the PCI ROM BIOS/CMOS SETUP configuration set
properly. You have to set it to allow shared memory access
from the ISA bus for the memory region that your card is trying
to use. If you can't figure out which settings are applicable
then ask your supplier or local computer guru. For AMI BIOS,
there is usually a "Plug and Play" section where there will
be an ``ISA Shared Memory Size'' and ``ISA Shared Memory Base''
settings. For cards like the wd8013 and SMC Ultra, change the
size from the default of `Disabled' to 16kB, and change the base
to the shared memory address of your card.
Do a cat /proc/interrupts
.
A running total of the number of interrupt events your
card generates will be in the list given from the above.
If it is zero and/or doesn't increase when you try to use
the card then there is probably a physical interrupt
conflict with another device installed in the computer
(regardless of whether or not the other device has a
driver installed/available).
Change the IRQ of one of the two devices to a free IRQ.
Werner Almesberger has been working on ATM support for linux. He has been working with the Efficient Networks ENI155p board ( Efficient Networks) and the Zeitnet ZN1221 board ( Zeitnet).
Werner says that the driver for the ENI155p is rather stable, while the driver for the ZN1221 is presently unfinished.
Check the latest/updated status at the following URL:
Is there any gigabit ethernet support for Linux?
Yes, there are currently several. One of the prominent Linux network developers rated the performance of the cards with linux drivers as follows: 1) Intel E1000, 2) Tigon2/Acenic, 3) SysKonnect sk-98xx, 4) Tigon3/bcm57xx. This was as of March 2002 and subject to change of course.
Is there FDDI support for Linux?
Yes. Larry Stefani has written a driver for v2.0 with Digital's DEFEA (FDDI EISA) and DEFPA (FDDI PCI) cards. This was included into the v2.0.24 kernel. Currently no other cards are supported though.
Will Full Duplex give me 20MBps? Does Linux support it?
Cameron Spitzer writes the following about full duplex 10Base-T cards: ``If you connect it to a full duplex switched hub, and your system is fast enough and not doing much else, it can keep the link busy in both directions. There is no such thing as full duplex 10BASE-2 or 10BASE-5 (thin and thick coax). Full Duplex works by disabling collision detection in the adapter. That's why you can't do it with coax; the LAN won't run that way. 10BASE-T (RJ45 interface) uses separate wires for send and receive, so it's possible to run both ways at the same time. The switching hub takes care of the collision problem. The signalling rate is 10 Mbps.''
So as you can see, you still will only be able to receive or transmit at 10Mbps, and hence don't expect a 2x performance increase. As to whether it is supported or not, that depends on the card and possibly the driver. Some cards may do auto-negotiation, some may need driver support, and some may need the user to select an option in a card's EEPROM configuration. Only the serious/heavy user would notice the difference between the two modes anyway.
If you spent the extra money on a multi processor (MP) computer then
buy a good ethernet card as well. For v2.0 kernels it wasn't really
an issue, but it definitely is for v2.2. Most of the older
non-intelligent (e.g. ISA bus PIO and shared memory design) cards
were never designed with any consideration for use on a MP machine.
The executive summary is to buy an intelligent modern design
card and make sure the driver has been written (or updated) to
handle MP operation. (The key words here are `modern design' - the
PCI-NE2000's are just a 10+ year old design on a modern bus.)
Looking for the text spin_lock
in the driver source is a good
indication that the driver has been written to deal with MP operation.
The full details of why you should buy a good card for MP use (and
what happens if you dont) follow.
In v2.0 kernels, only one processor was allowed `in kernel' (i.e. changing kernel data and/or running device drivers) at any given time. So from the point of view of the card (and the associated driver) nothing was different from uni processor (UP) operation and things just continued to work. (This was the easiest way to get a working MP version of Linux - one big lock around the whole kernel only allows one processor in at a time. This way you know that you won't have two processors trying to change the same thing at the same time!)
The downside to only allowing one processor in the kernel at a time was that you only got MP performance if the running programs were self contained and calculation intensive. If the programs did a lot of input/output (I/O) such as reading or writing data to disk or over a network, then all but one of the processors would be stalled waiting on their I/O requests to be completed while the one processor running in kernel frantically tries to run all the device drivers to fill the I/O requests. The kernel becomes the bottleneck and since there is only one processor running in the kernel, the performance of a MP machine in the heavy I/O, single-lock case quickly degrades close to that of a single processor machine.
Since this is clearly less than ideal (esp. for file/WWW servers, routers, etc.) the v2.2 kernels have finer grained locking - meaning that more than one processor can be in the kernel at a time. Instead of one big lock around the whole kernel, there are a lot of smaller locks protecting critical data from being manipulated by more than one processor at a time - e.g. one processor can be running the driver for the network card, while another processor is running the driver for the disk drive at the same time.
Okay, with that all in mind here are the snags: The finer locking
means that you can have one processor trying to send data
out through an ethernet driver while another processor tries to
access the same driver/card to do something else (such as get the
card statistics for cat /proc/net/dev
). Oops - your card
stats just got sent out over the wire, while you got data for
your stats instead. Yes, the card got confused by being asked
to do two (or more!) things at once, and chances are it crashed
your machine in the process.
So, the driver that worked for UP is no longer good enough - it needs to be updated with locks that control access to the underlying card so that the three tasks of receive, transmit and manipulation of configuration data are serialized to the degree required by the card for stable operation. The scary part here is that a driver not yet updated with locks for stable MP operation will probably appear to be working in a MP machine under light network load, but will crash the machine or at least exhibit strange behaviour when two (or more!) processors try to do more than one of these three tasks at the same time.
The updated MP aware ethernet driver will (at a
minimum) require a lock
around the driver that limits access at the entry points
from the kernel into the driver to `one at a time please'.
With this in place, things will be serialized so that the
underlying hardware should be treated just as if it was being
used in a UP machine, and so it should be stable. The downside
is that the one lock around the whole ethernet driver has
the same negative performance implications that having one big
lock around the whole kernel had (but on a smaller scale) - i.e.
you can only have one processor dealing with the card
at a time.
[Technical Note: The performance impact may also include
increased interrupt latencies if the locks that need to be
added are of the irqsave
type and they are held
for a long time.]
Possible improvements on this situation can be made in two ways. You can try to minimize the time between when the lock is taken and when it is released, and/or you can implement finer grained locking within the driver (e.g. a lock around the whole driver would be overkill if a lock or two protecting against simultaneous access to a couple of sensitive registers/settings on the card would suffice).
However, for older non-intelligent cards that were never designed with MP use in mind, neither of these improvements may be feasible. Worse yet is that the non-intelligent cards typically require the processor to move the data between the card and the computer memory, so in a worst case scenario the lock will be held the whole time that it takes to move each 1.5kB data packet over an ISA bus.
The more modern intelligent cards typically move network data directly to and from the computer memory without any help from a processor. This is a big win, since the lock is then only held for the short time it takes the processor to tell the card where in memory to get/store the next network data packet. More modern card designs are less apt to require a single big lock around the whole driver as well.
As of v2.0, only the 3c509, depca, de4x5, pcnet32, and all the 8390 drivers (wd, smc-ultra, ne, 3c503, etc.) have been made `architecture independent' so as to work on the DEC Alpha CPU based systems. Other updated PCI drivers from Donald's WWW page may also work as these have been written with architecture independence in mind.
Note that the changes that are required to make a driver architecture independent aren't that complicated. You only need to do the following:
-multiply all jiffies
related values by HZ/100 to account
for the different HZ value that the Alpha uses.
(i.e timeout=2;
becomes timeout=2*HZ/100;
)
-replace any I/O memory (640k to 1MB) pointer dereferences with the appropriate readb() writeb() readl() writel() calls, as shown in this example.
- int *mem_base = (int *)dev->mem_start; - mem_base[0] = 0xba5eba5e; + unsigned long mem_base = dev->mem_start; + writel(0xba5eba5e, mem_base);
-replace all memcpy() calls that have I/O memory as source or
target destinations with the appropriate one of
memcpy_fromio()
or memcpy_toio()
.
Details on handling memory accesses in an architecture
independent fashion are documented in the file
linux/Documentation/IO-mapping.txt
that comes
with recent kernels.
For the most up to date information on Sparc stuff, try the following URL:
Note that some Sparc ethernet hardware gets its MAC address
from the host computer, and hence you can end up with multiple
interfaces all with the same MAC address. If you need to
put more than one interface on the same net then use the
hw
option to ifconfig
to assign unique MAC address.
Issues regarding porting PCI drivers to the Sparc platform are similar to those mentioned above for the AXP platform. In addition there may be some endian issues, as the Sparc is big endian, and the AXP and ix86 are little endian.
There are several other hardware platforms that Linux can run on, such as Atari/Amiga (m68k). As in the Sparc case it is best to check with the home site of each Linux port to that platform to see what is currently supported. (Links to such sites are welcome here - send them in!)
Can I link 10/100BaseT (RJ45) based systems together without a hub?
You can link 2 machines, but no more than that, without extra devices/gizmos, by using a crossover cable. Some newer fancy autonegotiaton cards may not work on a crossover cable though. 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.
I get a bunch of `SIOCSIFxxx: No such device' messages at boot, followed by a `SIOCADDRT: Network is unreachable' What is wrong?
Your ethernet device was not detected at boot/module insertion
time, and when
ifconfig
and route
are run, they have no device
to work with. Use dmesg | more
to review the
boot messages and see if there are any messages about
detecting an ethernet card.
I get `SIOCSFFLAGS: Try again' when I run `ifconfig' -- Huh?
Some other device has taken the IRQ that your ethercard
is trying to use, and so the ethercard can't use the IRQ.
You don't necessairly need to reboot to resolve this, as
some devices only grab the IRQs when they need them and
then release them when they are done. Examples are some
sound cards, serial ports, floppy disk driver, etc. You
can type cat /proc/interrupts
to see which interrupts
are presently in use. Most of the
Linux ethercard drivers only grab the IRQ when they are
opened for use via `ifconfig'. If you can get the other
device to `let go' of the required IRQ line, then you
should be able to `Try again' with ifconfig.
When I run ifconfig with no arguments, it reports that LINK is UNSPEC (instead of 10Mbs Ethernet) and it also says that my hardware address is all zeros.
This is because people are running a newer version of the `ifconfig' program than their kernel version. This new version of ifconfig is not able to report these properties when used in conjunction with an older kernel. You can either upgrade your kernel, `downgrade' ifconfig, or simply ignore it. The kernel knows your hardware address, so it really doesn't matter if ifconfig can't read it.
You may also get strange information if the ifconfig
program you are using is a lot older than the kernel you are
using.
When I run ifconfig with no arguments, it reports that I have a huge error count in both rec'd and transmitted packets. It all seems to work ok -- What is wrong?
Look again. It says RX packets
big number PAUSE
errors 0
PAUSE dropped 0
PAUSE overrun 0
.
And the same for the TX
column.
Hence the big numbers you are seeing are the total number of
packets that your machine has rec'd and transmitted.
If you still find it confusing, try typing
cat /proc/net/dev
instead.
/dev/
for EthercardsI have /dev/eth0 as a link to /dev/xxx. Is this right?
Contrary to what you have heard, the files in /dev/* are not used.
You can delete any /dev/wd0, /dev/ne0
and similar entries.
How do I get access to the raw ethernet device in linux, without going through TCP/IP and friends?
int s=socket(AF_INET,SOCK_PACKET,htons(ETH_P_ALL));
This gives you a socket receiving every protocol type.
Do recvfrom()
calls to it and it will fill the sockaddr
with device type in sa_family and the device name in the
sa_data array. I don't know who originally invented
SOCK_PACKET for Linux (its been in for ages) but its superb stuff.
You can use it to send stuff raw too via sendto()
calls.
You have to have root access to do either of course.