1. Make sure there's a spare partition on the hard drive large enough to
    hold FreeBSD. The partition type doesn't matter, although it's a good
    idea not to select HFS.

 2. The initial install will have to be done with a network boot, or from a

    a) Network boot

	Set up a DHCP/TFTP/NFS server, with boot/loader in the directory that 
	will be served by TFTP. Untar the distribution in a directory that
	can be accessed with NFS, and put the tbz2 file in the root dir. The 
	entry that I am using for dhcpd.conf is

	  option subnet-mask;
	  option broadcast-address;
	  option routers;

 	  host emacfreebsd {
            hardware ethernet 00:03:93:A4:36:FE;
            option host-name "emacfbsd";
            always-reply-rfc1048 on;
            option root-path "";
            option routers;
            filename "loader";

	At the OpenFirmware prompt on the Mac (get to this with
	cmd-option-o-f at boot time):

	  0 > boot enet:,loader

	Hit the space bar when the loader is pausing to get to the loader 
	prompt, and then boot single-user

	  0K boot -s

	At the mountroot> prompt, enter 'nfs' to get to the single-user prompt.

     b) CDROM boot

	Create a directory tree with the bare minimum of files, extracted from
        the tarball, and the tarball itself, in the root directory. Here's
	what I used as a minimal tree:


	Build an ISO with the following options, required for OFW to read 
	the CD:

	   mkisofs -o  -hfs -part -l -J -r -L 

	With the CD in the mac drive (this can be done at the OpenFirmware 
	prompt on tray loaders with '0 > eject cd'), and at the Ofw prompt, 
	first verify that the CD can be read correctly

	0 > dir cd:,\
		 1/ 1/ 4  0: 0: 0  bin
		 1/ 1/ 4  0: 0: 0  boot
	   8192 10/24/ 3 11:24:35  Desktop%20DB
	      0 10/24/ 3 11:24:35  Desktop%20DF
	 	 1/ 1/ 4  0: 0: 0  dev
		 1/ 1/ 4  0: 0: 0  etc
		 1/ 1/ 4  0: 0: 0  mnt
        50878857 10/23/ 3 17:48:55  ppc-root-102303.tbz2
		 1/ 1/ 4  0: 0: 0  sbin
		 1/ 1/ 4  0: 0: 0  usr ok

	Now boot the loader, and break before it boots the kernel

	  0 > boot cd:,\boot\loader

	( hit space bar )

	  0 > boot -s

 	I've found that on the G4 eMac, the loader gets a boot device that 
	it doesn't understand, and spits out a whole lot of 'method  
	not found; ihandle=0 phandle=0'. Don't be too worried about this, 
	and at the loader prompt, enter:

	  OK set currdev=cd:,
	  OK boot -s

	At the mountroot> prompt, enter 'cd9660:acd0' to get to the 
	single-user prompt.

 3. At the shell in single-user mode:

	Have a look at the partitions just to make sure you will be using 
	the correct one. Here's the layout on my eMac:

	# /sbin/sysctl -b kern.geom.conftxt
	0 DISK ad0 40971571200 512 hd 16 sc 63
	1 APPLE ad0s14 5362777088 512 i 13 o 35608794112 n Home file system ty Apple_UNIX_SVR2
	1 APPLE ad0s13 8194153472 512 i 12 o 27414640640 n Eschatology 1 ty Apple_UNIX_SVR2
	1 APPLE ad0s12 8194153472 512 i 11 o 19220487168 n Swap ty Apple_UNIX_SVR2
	1 APPLE ad0s11 5414712320 512 i 10 o 13805774848 n A/UX Root ty Apple_UNIX_SVR2
	1 APPLE ad0s10 33554432 512 i 9 o 13772220416 n untitled 2 ty Apple_HFS
	1 APPLE ad0s9 13771429888 512 i 8 o 790528 n untitled ty Apple_HFS
	1 APPLE ad0s8 262144 512 i 7 o 528384 n Patch Partition ty Apple_Patches
	1 APPLE ad0s7 262144 512 i 6 o 266240 n Macintosh ty Apple_Driver_IOKit
	1 APPLE ad0s6 102400 512 i 5 o 163840 n Macintosh ty Apple_FWDriver
	1 APPLE ad0s5 37888 512 i 4 o 125952 n Macintosh ty Apple_Driver_ATA
	1 APPLE ad0s4 27648 512 i 3 o 98304 n Macintosh ty Apple_Driver_ATA
	1 APPLE ad0s3 37888 512 i 2 o 60416 n Macintosh ty Apple_Driver43
	1 APPLE ad0s2 27648 512 i 1 o 32768 n Macintosh ty Apple_Driver43
	1 APPLE ad0s1 32256 512 i 0 o 512 n Apple ty Apple_partition_map

	In this case, I'm going to dump the filesystem on to ad0s13. First, 
	newfs and mount the partition

	  # /sbin/newfs /dev/ad0s13
	  # /sbin/mount /dev/ad0s13 /mnt

	And now untar the distro

	  # cd /mnt
	  # /usr/bin/tar xjvf /ppc-root-102303.tbz2 .

	(this will take quite a while. I have found that the dreaded 
	 'ad0: device write underrun' has sometimes during this step - it 
	 pretty much means having to reboot and do the newfs/untar again :-(

	USB keyboard users can ctl-alt-del (aka ctl-option-del) to reboot 
	at this point. Otherwise, shutdown -h now.

	Another known bug: the system rarely shuts down cleanly. After 
	'uptime' has been displayed, it's time for a power cycle or 
	programmer's switch halt :-(

 4. It's time to boot the new system. However, here's the catch: OpenFirmware 
    doesn't understand UFS, so the loader has to live somewhere that OFW can 
    access it. This can be over the network as in step 2 a), or it can live 
    in a HFS+ partition that has MacOS/OSX installed. Here's how to boot when 
    loader is copied into the hard drive folder on both OS9 and OSX. Note 
    that loader can't understand HFS+, so it can't read the partition it was 
    booted from and fails (but that's OK).

      0 > boot hd:,\loader
      can't load 'kernel'

      0 > set currdev=hd:13     (or whatever partition was used)
      0 > boot -s

   At the mountroot> prompt, enter 'ufs:ad0s13'.

 5. It's now time to edit some of the system files so that multi-user can be

    a) Mount the root filesystem read-write so that vi can use tmp files

     # /sbin/mount -u -rw /dev/ad0s13 /

    b) Now set the terminal type to be something screen oriented, unlike the
       primitive type for /dev/console.

    # export TERM=cons25    (for those lucky enough to have syscons)


    # export TERM=ansi  (for OpenFirmware cons users. vt100 seems to work OK 
                          as well, but expect to be redrawing a lot in vi).

    c) Create an fstab

         /dev/ad0s13 / ufs rw 0 0

    d) Create /etc/rc.conf. Here's what I've been using. Enabling sendmail was
	causing a kernel hang at one stage.

	ifconfig_gem0="inet  netmask"

        # Not really necessary, but handy

	# seems to hang the system at boot otherwise :-(

	# shut up debug

    e) I usually enable telnet/ftp in inetd.conf, and enable secure access 
       to pty's

    f) Create the initial password databases. Since I cross-built on x86, the 
       build process creates little-endian db files, so they have to be 
       re-created with native byte-order (at least that's what I think is 

	# /usr/sbin/pwd_mkdb /etc/master.passwd

     g) It's ready to go multi with a ctl-D !