Gentoo on the 4780

I'm attempting a dual-boot Windows XP Pro/Gentoo 1.4 installation on my new Sager 4780. I'm expecting a snag or two (and, in fact, have already hit a couple), so I figured I'd document the process here.

The first step was booting from a Gentoo 1.4 LiveCD. Right off the bat, I hit the same snags some other people have reported on other Sager models -- some kernels would lock up, some would blank the screen, and others wouldn't recognize the network card.

What worked for me was booting with the "acpi" kernel. That doesn't bring the system up with hyperthreading enabled, but it does get me a running system with a working internal network card. Another option I tried that worked for me was booting "smp-nofb doacpi dopcmcia." With this kernel I could get the system up and running with hyperthreading, but the internal network card wouldn't work -- the "dopcmcia" part was so I would get support for a 3Com network card I'd popped into the PCMCIA slot ('cuz installing Gentoo with no network is a b****). It's also not as pretty as the console with the "acpi" kernel.

I'm currently in the bootstrap stage using the "acpi" kernel (I set USE="-java" first to avoid having to compile the GNU Java compiler), and everything appears to be going well so far. My plan when that's done is to build a 2.4 kernel with support for as much of my hardware as I can pull off. Assuming I get that far, I'll try to take the plunge into a 2.6 kernel.

Current deviations from the /etc/make.conf file (I plan to use GNOME on the machine, ALSA for sound, and include DVD and CD-writer support):

USE="X gtk2 gnome alsa dvd cdr -kde -qt"
CHOST=i686-pc-linux-gnu
CFLAGS="-O3 -march=pentium4 -funroll-loops -fomit-frame-pointer"
MAKEOPTS="-j3"

Edited to add:

Bootstrap started at approximately 10:40am, completed 1:40pm
"emerge system" started at 1:43pm

My guess is that it would've gone faster if I'd used an SMP kernel to enable hyperthreading -- without that, my "-j3" in the MAKEOPTS is probably slowing things down somewhat. I've also got the 4500 RPM 40GB drive... considering the amount of disk reads and writes involved in compiling this stuff, that's probably also a factor.

The reason I was thinking of going with a 2.4 kernel first is because... well, this isn't the first time I tried this. I had attempted using test sources (using ARCH=~i386 in make.conf) and a 2.6.1 kernel on my first go-around, and ended up with a kernel that would hang evey time I tried to boot it, after initializing the PCI devices.

I'll be building my kernel soon, from the looks of it -- if I'm feeling brave I may give 2.6.1 another shot.

While I'm here -- it looks like mirrorselect on the 1.4 LiveCD may be broken. My "emerge system" is taking forever, since using "mirrorselect -a -s4 -o >> /mnt/gentoo/etc/make.conf" had added MIRRORS="" to the end of the file, so emerge was avoiding the Gentoo mirrors altogether, and kept dying trying to download files from only their original sites. I must have manually restarted it a good 10 times before I figured that one out.

["emerge system" completed at 3:23pm]

OK -- I'm feeling brave. I've got a 2.6.1 kernel compiling right now. To keep from becoming overwhelmed with all the kernel options, I'm keeping it relatively stripped for now, and I'll add modules for hardware that doesn't work right off the bat.

Wow... it finished compiling already. I never cease to be amazed at how much faster this thing is than my old 200MHz IBM.

Now on to compiling the ATI drivers... That should keep it busy for a while.

Oh -- one more thing I discovered... booting with the "acpi" kernel from the LiveCD hangs if the internal modem is enabled in the 4780's BIOS. I had previously turned it off and didn't remember. As soon as I enabled it, the machine would no longer boot with that kernel -- it'd hang at the hardware discovery part of the boot process.

I'll wait and see if I can reenable it once I get everything else going.

Got everything ready, set up grub for dual boot, crossed my fingers, and rebooted...

Once again, I've got a kernel that won't boot. It would lock up immediately after loading the ohci1394 (for Firewire) module. I rebooted, reconfigured the kernel to (for now) disable Firewire support, and rebooted again.

This time it got as far as loading ehci_hcd (USB 2.0) before locking up. Back to the kernel config to disable USB for now. Looks like I'm really going to have to take this one piece at a time...

[A few minutes later]

Hmmm... Now it gets to the point of loading the r8169 (Ethernet) module, and locks up again. It appears something's screwed up to the point that loading any module will lock the box up. Maybe it wasn't a problem with Firewire and USB after all...

[Yet another kernel rebuild and reboot later...]

Built the ethernet driver into the kernel rather than using a module...

The last thing displayed on the screen is "Enabling device 0000:00:0a.0 (0015 -> 0017)" before the machine locks up again. An "lspci" shows that as the Ethernet controller.

I'll admit I'm getting just a *bit* frustrated by this whole thing.

That's pretty much the point I'm at now... I still can't seem to get past the kernel hanging when the network driver loads, though. I'm playing around a bit with some of the kernel options o see if I can find something somewhere that makes a difference. I'm also building a 2.4.22 kernel, to see if I get the same results.

Yep -- Gentoo merges 'em when you download the 2.6.1 source.

Interestingly, I finished building my 2.4.22 kernel, and it does exactly the same thing -- it locks up when loading the network driver. The card's still working; it loads up fine with the LiveCD kernel, it's just my kernels that lock it up.

Now, if I could just figure out what the difference between the two of them is, I might have a chance of getting this working.

Well, that's pretty much exactly what I did... I built two stripped-down kernels: one 2.6.1, and one 2.4.22. Both boot fine, but lock up as soon as I add the network driver (either as a module, or compiled-in).

I'm beginning to think it may not be a problem with the kernels themselves, per se, but maybe with the compiler toolchain generating bad code. The downside here is that the only way I can test that out is by recompiling things with less aggressive optimization flags or something. That's going to take some time.

For what it's worth, I also tried making a 2.6.1 genkernel, and it does the same thing the other kernels do. Since I know 2.4 kernels *can* work on this machine (the network driver on the LiveCD doesn't lock it up), it seems to be something in the way this machine builds the kernel and modules.

Unfortunately, I have to actually use this beast during the work week, so it may be a little time before I can try this out again.

I learned something new today. When Mom and Dad used to tell me that ignoring my problems wouldn't make them go away, they were sadly mistaken.

I slipped up at work today and selected the wrong option from my GRUB menu this morning, and accidentally booted my 2.4.22 kernel (with the network card module set up to load at boot)... and it worked. The machine didn't hang, and the network came up.

I have no rational explanation for this, but I can't say I'm unhappy about it.

I haven't had a chance to see if the 2.6.1 kernel works too, yet, but if it does I'll be a happy guy.

Fine by me... I think I'm just about to get to the good stuff.

I'm typing this in in Lynx under a 2.6.1 kernel. GNOME is still emerging at this point, and once that's done I'll really start digging into what works and what doesn't. So far, though, the built-in NIC is working great, and the console with the Gentoo bootsplash graphics looks pretty nice on the wide screen.

I've got a guess as to why it suddenly started working... As I mentioned before, I'd disabled the modem in the BIOS. Just to be safe, rather than just reenabling the modem, I reset the BIOS back to default values. The only change I made after that was to bump the AGP speed from 4x to 8x, and to reset the PNP configuration. My guess is that some other setting I changed was causing the freezes. I'll have to do some playing around to see if I can break it again, just to find out which setting it was.

At this point, though, I can happily say that Gentoo does, indeed, boot and run on the 4780.