karmic installer fails to detect Intel 82567 network card

[Mathieu Benoit]

Binary package hint: debian-installer

When attempting to install karmic over the network, the network hardware was not detected and the install failed. This happened on two machines:
* Toshiba Tecra M10
* Toshiba Tecra A10
These machines have Intel 82567v and 82567lm network cards.
Executing "ip link" from the command line did not show the network interface.

I do more test, we don't have this error on Jaunty.

[Leann Ogasawara (leannogasawara)]

Hi Mathieu,

I'm reassigning this to the kernel for now. I've asked fader to attach the relevant 'sudo lspci -vnvn' information for those network cards. I basically need to know the pci id's. Thanks.

[Brian Murray (brian-murray)]

The pci ids are likely:

8086:10cb 82567V Gigabit Network Connection
8086:10f5 82567LM Gigabit Network Connection

[Mathieu Benoit (mathben)]

Hello Leann, this is information about the network cards! I attach all information on a file!

[Leann Ogasawara (leannogasawara)]

I'm only seeing the information for the following in the attached file:

00:19.0 Ethernet controller [0200]: Intel Corporation 82567LM Gigabit Network Connection [8086:10f5] (rev 03)

This confirms one of the device id's Brian suspected. Can you confirm the other id he posted by attaching the sudo lspci -vnvn for the system with the other 82567V card?

[Mathieu Benoit (mathben)]

Sorry, I forget the other network card...
00:19.0 Ethernet controller [0200]: Intel Corporation 82567V Gigabit Network Connection [8086:10cb] (rev 03)

[Leann Ogasawara (leannogasawara)]

Thanks Mathiue. I notice both these cards require the e1000e driver. This driver looks to be enabled as a module in Karmic and is also in the corresponding nic-modules files so that it should be passed and available to the installer.

ogasawara@emiko:~/ubuntu-karmic/debian/config$ grep -rn "CONFIG_E1000E" *
config.common.ports:765:CONFIG_E1000E=m
config.common.ubuntu:866:CONFIG_E1000E=m

ogasawara@emiko:~/ubuntu-karmic/debian/d-i/modules$ grep "e1000e" nic-modules
e1000e ?

So I'm a bit puzzled as to why this would be failing. Could you try to install with the latest daily builds and if the issue remains, attach dmesg output?

[Mathieu Benoit (mathben)]

I will ask Ronald (fader in launchpad) to change the configuration. I will give you feed back in the following days.

[Ronald McCollam (fader)]

Leann,

This is occurring during the install, so it is a part of the install environment rather than the installed system. Would that make a difference as to what modules are available to the kernel at that time?

If it would be helpful to you I can get e.g. the syslog from the install process up to that point, though it might take me a day or so as I'm not physically located near the system in question and it has no network connection during install.

[Leann Ogasawara (leannogasawara)]

Hi Ronald,

If you could get the syslog and attach it here that would be great. Having the e1000e driver in the nic-modules files should make the driver available to the installer so I'd be interested to see if there are any messages in the syslog which may help figure out why this is now failing.

[Leann Ogasawara (leannogasawara)]

Each syslog is showing the following:

Aug 14 13:47:23 kernel: [ 2.372905] e1000e: Intel(R) PRO/1000 Network Driver - 1.0.2-k2
Aug 14 13:47:23 kernel: [ 2.373026] e1000e: Copyright (c) 1999-2008 Intel Corporation.
Aug 14 13:47:23 kernel: [ 2.373267] alloc irq_desc for 20 on node -1
Aug 14 13:47:23 kernel: [ 2.373269] alloc kstat_irqs on node -1
Aug 14 13:47:23 kernel: [ 2.373275] e1000e 0000:00:19.0: PCI INT A -> GSI 20 (level, low) -> IRQ 20
Aug 14 13:47:23 kernel: [ 2.373374] e1000e 0000:00:19.0: setting latency timer to 64
Aug 14 13:47:23 kernel: [ 2.373465] alloc irq_desc for 29 on node -1
Aug 14 13:47:23 kernel: [ 2.373466] alloc kstat_irqs on node -1
Aug 14 13:47:23 kernel: [ 2.373476] e1000e 0000:00:19.0: irq 29 for MSI/MSI-X
Aug 14 13:47:23 kernel: [ 2.438588] 0000:00:19.0: 0000:00:19.0: The NVM Checksum Is Not Valid
Aug 14 13:47:23 kernel: [ 2.458805] e1000e 0000:00:19.0: PCI INT A disabled
Aug 14 13:47:23 kernel: [ 2.494284] yenta_cardbus 0000:06:0b.0: CardBus bridge found [1179:0001]
Aug 14 13:47:23 kernel: [ 2.504237] e1000e: probe of 0000:00:19.0 failed with error -5

The interesting line seems to be:

Aug 14 13:47:23 kernel: [ 2.438588] 0000:00:19.0: 0000:00:19.0: The NVM Checksum Is Not Valid

Which seems to remind me of the notorious e1000e driver bug we saw a while back which ate people's hw:

http://lkml.org/lkml/2008/9/25/510
http://bugzilla.kernel.org/show_bug.cgi?id=11382
https://bugs.edge.launchpad.net/ubuntu/+source/linux/+bug/263555
https://bugzilla.redhat.com/show_bug.cgi?id=459202

The patches that are in the current Karmic kernel should prevent this from happening in the future, but does not fix already affected/damaged hw. What I'm concerned is you mentioned this was working in Jaunty which should have already had these patches applied. And it seems these systems were working fine with early Karmic releases as well according to the test results:

https://certification.canonical.com/hardware/200810-880
https://certification.canonical.com/hardware/200810-881

Can you dump your eeprom using ethtool -e

I'm going to raise this to the rest of the kernel team. Thanks.

[Leann Ogasawara (leannogasawara)]

I've also sent an email to Jesse, the kernel dev noted in the above LKML threads and bugzilla bug report.

[Jesse Brandeburg (jesse-brandeburg)]

I suspect that this patch will fix the issue, seems that since they were able to boot the old OS and dump the eeprom that the eeprom is not actually corrupt.

http://git.kernel.org/?p=linux/kernel/git/davem/net-2.6.git;a=commitdiff;h=148675a7b2061b5a5eb194530b7c4d8de1f2887e;hp=373a88d78be540c1331ea5adcb76610dddcb008b
patch above is also attached.

This patch is currently in net-2.6 and should be pushed in time to be in 2.6.31-rc7

The other patch to try (applies before the above) is:
http://git.kernel.org/?p=linux/kernel/git/davem/net-2.6.git;a=commitdiff;h=373a88d78be540c1331ea5adcb76610dddcb008b

[Jesse Brandeburg (jesse-brandeburg)]

probably the same bug as
http://bugzilla.kernel.org/show_bug.cgi?id=13914

[Leann Ogasawara (leannogasawara)]

Sweet, thanks Jesse.

Ronald, Mathieu - I've built a test kernel with the suggested patch applied. Please test and let us know your results:

http://people.canonical.com/~ogasawara/lp404264/

[Mathieu Benoit (mathben)]

I have a good news!
I install Karmic with from installation CD.
After a complete installation, I run ifconfig and "eth0" doesn't appear.

I install Leann's patch (http://people.canonical.com/~ogasawara/lp404264/).
After a reboot, "eth0" appear on ifconfig and i can surf on the net!

I do this step for the two laptop and it's working!

Do you need more information about the laptop?

[Leann Ogasawara (leannogasawara)]

Thanks for testing and the feedback Mathieu. That's all I needed to know. The patch that was tested is now in the upstream kernel. This will be pulled into Karmic automatically once we do the next rebase. Thanks.

ogasawara@emiko:~/linux-2.6$ git log 148675a7b2061b5a5eb194530b7c4d8de1f2887e
commit 148675a7b2061b5a5eb194530b7c4d8de1f2887e
Author: Bruce Allan <email address hidden>
Date: Fri Aug 7 07:41:56 2009 +0000

    e1000e: fix potential NVM corruption on ICH9 with 8K bank size

    The bank offset was being incorrectly calculated on ICH9 parts with a bank
    size of 8K (instead of the more common 4K bank) which would cause any NVM
    writes to be done on the wrong address after switching from bank 1 to bank
    0. Additionally, assume we are meant to use bank 0 if a valid bank is not
    detected, and remove the unnecessary acquisition of the SW/FW/HW semaphore
    when writing to the shadow ram version of the NVM image.

    Signed-off-by: Bruce Allan <email address hidden>
    Signed-off-by: Jeff Kirsher <email address hidden>
    Signed-off-by: David S. Miller <email address hidden>

[Leann Ogasawara (leannogasawara)]

Looks like this is now indeed in the Karmic kernel and should be available for testing with the most recent 2.6.31-8 kernel. Marking this Fix Released. Thanks.

ogasawara@yoji:~/ubuntu-karmic$ git log 148675a7b2061b5a5eb194530b7c4d8de1f2887e
commit 148675a7b2061b5a5eb194530b7c4d8de1f2887e
Author: Bruce Allan <email address hidden>
Date: Fri Aug 7 07:41:56 2009 +0000

    e1000e: fix potential NVM corruption on ICH9 with 8K bank size

    The bank offset was being incorrectly calculated on ICH9 parts with a bank
    size of 8K (instead of the more common 4K bank) which would cause any NVM
    writes to be done on the wrong address after switching from bank 1 to bank
    0. Additionally, assume we are meant to use bank 0 if a valid bank is not
    detected, and remove the unnecessary acquisition of the SW/FW/HW semaphore
    when writing to the shadow ram version of the NVM image.

    Signed-off-by: Bruce Allan <email address hidden>
    Signed-off-by: Jeff Kirsher <email address hidden>
    Signed-off-by: David S. Miller <email address hidden>

[Ronald McCollam (fader)]

We can confirm that the installer now sees the network device. Thanks folks!

[sanam (sanam79er)]

I installed Ubuntu 9.10 after installing Windows 7. My internet connection over windows is perfect, but I I cannot connect to the ethernet as well as wireless network on Ubuntu.
Laptop: Lenovo Thinpad T400s
ethernet controller: Intel 82567
Linux sanam-laptop 2.6.31-17-generic #54-Ubuntu SMP Thu Dec 10 16:20:31 UTC 2009 i686 GNU/Linux