The Case of the Mysterious Missing MAC Address

Coriander is dead. Turmeric is alive. What does this mean? All the content on sftp.knarrnia.com went away except for the PDFs and various other ebook formats which I was smart enough to grab off the RAID before it entirely keeled over. The uptime for coriander was ultimately 20 minutes tops before it keeled over. Not bad for a computer I built in 2003 and then rebuilt when Drexel had the Heat Wave of Death which caused me to request an extension on the finals. The box was two RAIDed 80GB IDE drives, running OpenSuSE I had installed as a desktop and later simply retired to serving up content to my Nook and XBOX.

Turmeric, however, is a first gen nVidia motherboard. And if this is people’s experience with nVidia, I am entirely, absolutely done with them as a motherboard maker. My network is pretty standard for home use. I have an honest to god Cisco router, a Cisco WAP, a comcast bullshit cable modem which is probably going to have a terrible accident so I can get one that works, and all the devices meander through those. The XBOX is UPnP permissions, nothing else does. Turmeric/Coriander had a MAC address reservation so they would come up, get the right IP, and then the cisco firewalls would pass traffic to them as needed. It worked swimmingly well until Turmeric refused to get the IP I had reserved for it. It would always get a different IP than reserved, but it would get the IP consistently. I racked my brain on this problem for a few hours and finally broke out ettercap to see WTF it was doing.

Turns out the first gen nVidia motherboards do something really stupid with DHCP. Actually lets rewind for a minute – they generally do really stupid stuff. This motherboard has hardware RAID also, but it only works for the SATA drives. IDE? Shit out of luck. To further add insult to injury, the bootp stuff for jumpstarting a box? Doesn’t work. Never figured that one out. Finally there’s only two default devices you get set in the BIOS. For the moment it’s CDROM and then the first drive in the RAID, but to actually do the install I had to change CDROM to USB after burning out an image to it. What the heck guys?

Now, I’ll save you the boring TCP spec – When the nVidia board comes up it actually sends a DHCP packet on its own which is nice. The problem here is the HLEN of the packet is… 0. Yup. Someone didn’t know what to put in the field, so they send 0. This causes the router (thank god) to respond to FF:FF:FF:FF:FF:FF, which while it’s not correct, works because it’s a broadcast packet. The adaptor (seems) to configure itself, then Linux does something goofy where it sees the adapter is configured, so it sends out a release/renew, which the router, apparently knowing the MAC address but having an entry for a bogus MAC, sends out a different IP (next in pool) for the correct MAC address.

The BIOS, of course, doesn’t have a way to disable this “convenience feature” and to add insult to injury, dmesg doesn’t work in Linux because the BIOS is doing something funky by itself. For right now I’m just ignoring it. But seriously nVidia, fix your shit.

nVidia fix for KDM not starting

If you’re like me and you updated your Linux desktop to find Xorg (because of nVidia) no longer played with XDM, KDM, or GDM then you’ve hit the hilarious ignoreABI bug.

For some people, editing their /etc/X11/xorg.conf works:

Section "ServerFlags"
	Option		"IgnoreABI" "True"
EndSection

Not me. No matter what flags I tried to pass in via xorg.conf, it wouldn’t go. Probably because sometimes the GUI is looking for the configuration, othertimes it’s not. According to /var/log/kdm.log it was, but it cowardly refused to honor that directive.

The fact that the ubuntu guys call it a “good driver” just means the typical ubuntu user has no idea what a good driver looks like.

Finally someone with half a brain ran into this on Xorg 1.5 and their new nVidia card. You can read their fix here, it’s basically what you would expect. Shim out /usr/bin/Xorg with a script, have the script call the other Xorg.0 executable and pass all the arguments with -ignoreABI. Incidentally something really strange is happening here because editing /etc/X11/xinit/xserverrc to add -ignoreABI to the args line doesn’t get passed to Xorg.

I Hate nVidia

I have a dirty confession – I’ve always liked ATI stuff except for when the GeForces first came out and they were cheap as heck. Buying two of them would buy you a high end 3DFX card or ATI card and it would still outperform them. Also these were the college days when AGP was still new and having two videocards meant one less thing to kill your PC. Then again this was drexel, and we had the Kelly Hall heat wave that year, and it killed my motherboard.

ATI fought and fought hard to get back to the top, and it was only after AMD bought out the last of the DEC stuff for really awesome 64bit support and then gobbled up ATI that things got good again. Frankly it was a great move since graphics are almost all math, so having a 64bit (or even 128bit) pipe with multipath and short-lines is just great.

The came the licensing wars.

Linus (correctly) said that kernel shims were OK so long as they’re open source. He’s no dummy, kernel shims let the kernel load blobs, but being open source they replace the linking and once you’ve got the linking objects you’re most of the way to having a driver since you can see what the card is being sent and you can see what the kernel is sending. Open source drivers followed, but some of the really exotic stuff only recently caught up.

nVidia has always, always been a pain in the ass in Linux. The shim wouldn’t build when it first came out and required users to edit the Makefile, certain gcc versions produced drivers which were slow or had unintended consequences depending on how they did memcpy and other low level functions. Installing nVidia was mostly a one way ticket to either kernel lock-in or building it by hand. To further add insult to injury, nVidia never offered a unified driver and always had three versions. Now this was OK up until recently – They kept a list of cards so you generally knew if you needed nv, nvidia-G01 or nvidia-G02. Now the bad news: nVidia has decided to drop updates for older cards. I realize they can’t update them forever, but what’s missing? Open sourcing the drivers.

ATI hasn’t really offered up any open source drivers, but they did offer unified drivers. Download one, build it, you win! The build process is pretty seamless. ATI hasn’t moved to quash open source drivers either, to the point where the open source drivers are so stable that they are now officially merged to MESA. If you’re wondering what MESA is, MESA provides openGL functionality to the system in a common package. To have drivers in it for a major manufacturer like ATI means you simply install MESA and 3D just works. No more diddling around with drivers, third party crap, and the ATI clock tray icon (unless you want to).

Now if you’re like me, you’re running OpenSuSE. You’re probably not like me but you might be running Linux. Windows users should have stopped reading six paragraphs ago. I upgraded to 11.3 from 11.1 (which I needed to run to hack the novell client from SuSE 10 into working because novell doesn’t even update their own stuff) and what broke? Oh, the nvidia drivers. Given that this is a work PC, I have no sway in my videocard. I went to fire up sax2 and I was told it was deprecated because of XOrg updating their autodetection routines. The new XOrg is nice, the new SuSE is nice, but with no new nVidia release my KDM login manager doesn’t work. Weirdly enough I can log in on the console and do a startx which does work, but it would be nicer to have a GUI running. (Then again having an ominous text console keeps the n00bs off my PC). After hacking on this most of the last few days, it’s definitely a problem in how nVidia does the initialization and it’s directly related to the fact I am running nvidia-G01. Way to go nVidia.

My laptop (ATI)? Runs great, and it’s a radeon mobility 600. Hardly new. Guess we know who’s videocard I’ll be buying in the future.