D945GCLF & Linux - Some Serious Head Bumping In Most Recent Kernels Re Ethernet

[mgrollman]

We are doing some prototype mobile stuff with the new Atom board, D945GCLF, using fairly recent versions of Linux. Most of the kernels we try to use (from 2.6.20 through 2.6.25) either crash the system on boot (modprobe errors related to the realtek Ethernet port) or else successfully start the system, but only operate the Ethernet port point sporadically -- most of the failures are warm-boot related. By building we the latest code, we can get past the OS crash problem, but still, the warm boot issues remains.

There are several somewhat hard-to-nail down kernel patches floating around to deal with the first set of issues, crashing the whole OS. But I have not seen anyone address the 2nd set. The Mandriva folks seem to have a fix, as do the Ubuntu people, not not necessarily for both issues, and not in a systemically implemented way.

Anyone got some ideas? I am not sure where is the right place to post, but since there did not appear to be anything anywhere on the Intel site about this neat little board crashing a very kind OS like Linux, all for a wee working driver, I figured someone ought to put something out there.

Full technical details are available with a Google of 'D945GCLF Linux bug' . One OK place to look is also http://www.nabble.com/-Bug-41198--NEW:-r8169-module-segfaults-on-Intel-Atom-D945GCLF-Mobo-p17591102.html.

Cheers,

- Michael

[James_M_Intel1]

I dont have a direct answer for you but a colleague of mine says hes used Ubuntu 7.04 just fine with this board. He also believe 7.10 works as well. He also provided a link to this tread which suggests a couple of work-arounds: use another network card or update to the 2.6.24-19-generic kernel. A poster also reports that Ubuntu Hardy 8.04.1 from June 20, 2008 works as well.

http://ubuntuforums.org/showthread.php?t=812817

[mgrollman]

Lester -- thx for the prompt feedback. I appreciate that some really new Ubuntu and other kernels are getting past the first critical error, the one that stops the OS from booting on the D945GCLF -- the mobprobe Ethernet crash one.

However, there is a 2nd level problem, also I suspect critical but harder to find, which appears to relate to an intermittent warm boot issue, that may be related to a capacitor-full-discharge issue near the Ethernet port, or it may be related to the driver leaving the port in an unknown state. Regardless, if someone has a clean read on which patches fix it, that would be good, as we like to build pristine kernels, to better control for software variability, that to start with a distro one. Any thoughts?

- Cheers, Michael

[James_M_Intel1]

Sorry. I don't have any more information about the 2nd level problem you described. Hopefully someone else who frequents this forum will.

[mgrollman]

To put some additional flesh on it, the first level issue, making the board not crash on boot, is dealt with here: http://lkml.org/lkml/2008/4/17/298. The short version is "When does r8169.c finds unknown MAC then kernel crash occurs because tp->pci_dev is NULL and is dereferenced."

Trouble is, this patch may stop the initial crash, but still 
does not seem to properly support the RTL8102E chip set included
with the D945GCLF board, in actual operation.

Since the board has only the single PCI slot, it is hard to justify
adding a network card, when there are other peripherals
that may need to be included (video in my case).