converted to 1.6 markup
|Deletions are marked like this.||Additions are marked like this.|
|Line 11:||Line 11:|
|Input can be sent through my [[http://www.calel.org/contact.html|contact page]].
|Line 59:||Line 57:|
|deb http://debian.calel.org/macpro etch main|| deb http://apt.debianchile.org/macpro etch main
You may want to install the archive key as well,
# gpg --keyserver hkp://subkeys.pgp.net --recv-keys EFD17969
# gpg --export EFD17969 | apt-key add -
By RicardoYanez. This page describes a method to install Debian on Apple Intel Mac Pro computers. It has been tested in two computer systems with a Quad-Core and dual Quad-core Xeon 5300-series "Clovertown" processors.
(If you are trying to install onto anything other than the primary disk drive (i.e. the one that Mac OS X boots from), then read the "Booting from a second disk" section below (comments on that section to MatthewVernon at his Debian address, please).)
If you have already tried to install Debian on Intel Mac Pro, you may have noticed the Debian Installer CD doesn't mount the CD-ROM properly. Eventually, no media is available to continue with the installation.
For the inpatient, the next section explains how to install Debian with a custom Debian Installer mini CD and custom Linux kernel in few steps.
In the "Details" section I explain how I compiled the custom kernel and how I made the custom Debian Installer CD.
Download the Debian Installer Mini CD for Intel Mac Pro (MD5SUM 023dadbea2bb5bc064d81b4fc62f184f), burn it to a CD, for example,
$ wodim -v -eject dev=/dev/cdrw -data mini.iso
Being a mini CD, it contains no .deb packages. You will need a network connection to install the base system.
Under Mac OS X, install rEFIt, the EFI boot menu for Intel Mac. Easiest is to download the Mac disk image, double click on the icon and run the installer.
Still under Mac OS X, open the CD tray, insert the Debian Installer Mini CD and reboot. In the boot menu choose the penguin CD.
Boot the installer in expert mode. Install as you would in any system.
Choose to partition manually, then set the root partition bootable flag on.
During the installation of the base system, choose 'none' for the kernel to install.
Choose to continue without boot loader.
Just before finishing the installation, open a virtual terminal, <ctrl>-<alt>-<f2> for instance, and run a shell. Change the root directory with,
# chroot /target
Install grub and initramfs-tools with,
# apt-get install grub initramfs-tools
# grub-install /dev/sda
to install GRUB on the first disk drive.
Edit /etc/apt/sources.list and add line,
deb http://apt.debianchile.org/macpro etch main
You may want to install the archive key as well,
# gpg --keyserver hkp://subkeys.pgp.net --recv-keys EFD17969 # gpg --export EFD17969 | apt-key add -
Run 'apt-get update' and install the custom kernel,
# apt-get install linux-image-2.6.20-2-macpro-amd64
Say no to abort kernel installation.
Run update-grub and create /boot/grub/menu.lst.
Go back to the installation terminal, <ctrl>-<alt>-<f1> and finish the installation.
In the boot menu choose the penguin disk.
After boot, you may want to install the kernel headers,
# apt-get install linux-headers-2.6.20-2-macpro-amd64
If you like to make Debian the first boot device, under Mac OS X, edit /efi/refit/refit.conf and uncomment option legacyfirst.
If you feel uncomfortable using my custom kernel, compile your own using the mactel-linux patches as described below.
(This part requires some familiarity with kernel compilation and debian packaging).
I have two 64-bit dual Quad-Core Xeon E5335 "Clovertown" machines at work. One is an HP server, the other an Apple Intel Mac Pro. Both are primarily used to run and develop Monte Carlo simulations in parallel, or otherwise.
I obviously had no trouble installing Debian Etch on the HP server. On the Mac, the first thing one needs to do is to install rEFIt under Mac OS X, open the CD tray to insert the installation CD, boot and choose the penguin.
The Debian installer boots, but it doesn't mount the CD-ROM. Eventually, no media is available to continue with the base-system installation. I tried the standard approaches given in several wikies on the subject, e.g. boot parameters like 'install noapic irqpoll acpi=force', and combinations thereof, but none really helped. I also tried a USB-key installation, only to learn Mac firmware doesn't support USB-key installations, at least in the machines I have.
Then a co-worker handed me an Ubuntu 7.04 Feisty server installation CD, which recognized the CD-ROM, mounted the CD, installed the base system, and went all the way to the finish line. After boot, the kernel would not load, though. I read several guides describing some very intricate recipes, which included the need of Live CDs and disk partition utilities. None of these approaches is really necessary, once you realize the only thing they are trying to do is to set the boot flag on the root partition.
This exercise made me realize the installer needs to attach the CD-ROM not as ATAPI IDE (ide-cd) as the Debian Installer does, but under SATA (not sure here why, nor if this statement is absolutely true.)
My plan of action was then to compile a custom kernel a la Ubuntu (which has the Mactel-linux patches included) that attaches the CD-ROM as the hardware requires, then make a new Debian Installer CD with it.
In short, kernel 2.6.20 is compiled with the mactel-linux patches, then debian-installer is used to create the mini installation image.
Custom kernel 2.6
I compiled the kernel in the HP server running Debian Etch. I used kernel version 2.6.20, downloaded from kernel.org. Untar the tarball in /usr/src.
Install subversion and get the mactel-linux kernel patches,
# svn co https://mactel-linux.svn.sourceforge.net/svnroot/mactel-linux mactel-linux
Go to trunk/kernel/mactel-patches-[kernel-version] and apply patches with,
# ./apply /usr/src/[linux-source-directory]
Install kernel-package and libncurses5-dev. Go the the kernel source directory and 'make menuconfig'.
This is the current .config file.
Before compiling, edit Makefile and remove the EXTRAVERSION introduced by the Mactel-linux patching. Leave it blank.
# make-kpkg --initrd --append-to-version=-1-macpro-amd64 --revision=2.6.20 kernel_image kernel_headers
Install the custom kernel. No need to boot it in what follows.
I'm repeating here some, if not most, of DebianInstaller/Modify/CustomKernel.
Install devscripts and kernel-wedge.
Unpack the source package linux-kernel-di-amd64-2.6,
# apt-get source linux-kernel-di-amd64-2.6
Go to the directory and edit kernel-versions to match the new version, 2.6.20-1-macpro-amd64. Comment out or delete the existing line.
Check for build dependencies with dpkg-checkbuilddeps, and install dependencies as needed.
If you have a GPG signature you can do 'debchange -i' to edit the changelog with your e-mail address and some comments. The directory name will change, reflecting the new version number. You will have to go to the parent directory, then back into the new directory. You may also want to change the Maintainer: field in debian/control.stub, and define the EMAIL and GNUPGHOME environment variables in the user's .bashrc, then source it.
Build the package with debuild, or 'debuild -rfakeroot' if not root.
If you have no GPG signature, build with 'debuild binary', or 'debuild -rfakeroot binary' if not root instead.
The building may fail due to missing kernel modules. Edit module listings in modules/amd64 needed be. I copied some of the include files into the directory and put a ? at the end of the missing module. See kernel-wedge documentation for more options. I had to delete some module-list files that generated empty udebs.
If you have a GPG signature, sign the files.
You should end up with a bunch of udebs in the parent directory.
Get the source,
# apt-get source debian-installer
Go to directory and check for dependencies.
Edit build/config/amd64.cfg and change KERNELVERSION. Uncomment monolithic image type.
Copy all kernel udebs into build/localudebs.
Go to build/ and build the monolithic image,
# make build_monolithic
The image is placed in dest/monolithic.
Now you have the CD image and the kernel image. Install as described in the installation section.
Booting from a second disk
rEFIt assumes that you have only one disk drive. If you try and install linux onto a secondary drive, you will probably have found that rEFIt lets you try and boot your newly-minted linux partition/drive, only for you to get a "Missing operating system" error message. This is actually a Syslinux error message. What happens is that rEFIt looks on the primary disk for an MBR record, fails to find one (obviously!), so sticks the syslinux MBR onto the primary disk, and tries to boot that. You can confirm that this has happened, with some runes like the following:
# dd if=/dev/sda of=/tmp/mbr bs=512 count=1 # file /tmp/mbr
Hopefully this rEFIt bug will be fixed, but in the mean-time, the best work-around is to install GRUB or LILO onto the MBR of the primary hard-disk, and use that to boot your partition. This means you'll end up having to select linux at the rEFIt menu, and then again in the GRUB or LILO menu, but you can always turn the timeout down very low in LILO Something like this in lilo.conf:
boot=/dev/sda root=/dev/sdb2 image=/boot/vmlinuz-2.6.20-2-macpro-amd64 label="Lin 2.6.20img0" initrd=/boot/initrd.img-2.6.20-2-macpro-amd64 read-only
Should do the trick.