Network Does Not Start Using be2net Driver on Lucid

[Joseph Salisbury]

An attempt is being made to image an HP Proliant BL 460c G7 that uses the be2net module. The latest Lucid netboot kernel and initrd that has module included is being used. However, the host refuses to bring the network up, using dhcp or static addresses.

This same procedure was performed using Maverick. This problem does not exist in Maverick and a network install can be performed successfully.

Some additional information was found performing tests with other kernels. The following specific test was done using the kernel and initrd for 2.6.32-25-generic to install. (However, the latest kernel in lucid-updates - 2.6.32.28 was tested as well and also failed to start the network).

The following test was performed was able to get the network to start:
1. Downloaded the kernel source for 2.6.33-7 from linux.org, also installed kernel-source-2.6.32 for the -25 kernel.

2. Copied drivers/net/benet to linux-source-2.6.32/drivers/net/benet. A change was needed in the Makefile to be 2.6.32.25-generic.

3. Performed rsync /usr/src/linux-headers-2.6.32-25-generic/include/ to /usr/src/linux-source-2.6.32/include.

4. The file /boot/config-2.6.32-25-generic to was copied to .config. and the following was run: make oldconfig, make prepare, make scripts and finally make modules M=drivers/net/benet. Which gave a compile time error.

5. This was tracked down to /usr/src/linux-source-2.6.32/include/linux/skbuff.h. Lines 1480 - 1488 of /usr/src/linux-2.6.33.7/include/linux/skbuff.h were copied into /usr/src/linux-source-2.6.32/inculde/linux/skbuff.h.

The module then compiled without error. It was placed into the initrd. After these changes, the host is able to DHCP and install lucid.

So whatever it is, changes made to this driver in 2.6.33+ seem to have fixed it.

However, even without the previous test, it is confirmed that the network works in Maverick but not in Lucid with the be2net driver using the same hardware.

[Joseph Salisbury (jsalisbury)]

Output from apport-bug

[Joseph Salisbury (jsalisbury)]

dmesg output

[Joseph Salisbury (jsalisbury)]

lspci output

[Tim Gardner (timg-tpi)]

I'm seeing some ACPI errors during the be2net startup which suggests MSI routing might be an issue. Can you boot the kernel with 'pci=nomsi' ? Otherwise it looks like eth0 and eth1 found a link.

[Joseph Salisbury (jsalisbury)]

Thanks, Tim. We will try booting with pci=nomsi and update the bug with the results.

[Joseph Salisbury (jsalisbury)]

@Tim, we tried a boot with pci=nomsi. However, there was no change, still the same issue.

[Joseph Salisbury (jsalisbury)]

@Tim, we also tried an install using the proposed 10.04.2 point release at http://cdimage.ubuntu.com/lucid/daily/current. However, this has the same issue.

Are there any other suggestions we may want to try?

Thanks,

Joe

[Tim Gardner (timg-tpi)]

It looks like the 2.6.32 driver simply doesn't support this new BladeEngine hardware. Would it be sufficient if I provided a linux-backports-module package that could be installed after the fact? The user would have to initially install using a USB ethernet dongle, but could then get native ethernet support after installing linux-backports-module-benet.

[Joseph Salisbury (jsalisbury)]

@Tim

This may not be the best solution since all of the servers affected by this live in a data center, in another building. PXE is used to boot and install Ubuntu over the network. Its not feasible to have people going to remote labs with other ethernet devices to plug into the blade chassis every time an install is needed.

If this were not on an LTS version, It would most likely be an acceptable solution. However since this is a LTS version 5 years would be a long time to use this solution.

[Tim Gardner (timg-tpi)]

Joseph - The 10.04.2 DVD has the Maverick kernel version as a boot and install option. The magnitude and number of changes required to update the Lucid in-kernel driver do not meet SRU requirements. Those appear to be your choices.

[Joseph Salisbury (jsalisbury)]

Thanks for the feedback, Tim.