You need a root filesystem for the to kernel mount at startup. There are many options, and the best one will generally depend on whether your system needs to be able to store persistent data (which must survive power cycling) in the field.
See: http://lists.linuxppc.org/listarcs/linuxppc-embedded/200003/msg00067.html.
Your best bet is likely to be the root file system from Hard Hat Linux. Extract all the RPMs matching *.noarch.rpm, and use the image in opt/hardhat/devkit/ppc/8xx/target as the root file system.
During development, the embedded system can NFS-mount its root file system from your file sever to provide a complete diskless Linux system. The file server need not be the same architecture as the embedded client. Answer "Y" to the kernel configuration questions regarding NFS client and root filesystem via NFS, and "make zImage". The embedded system will attempt to mount its root file system from the server as "/tftpboot/<ipaddress>", where <ipaddress> is it's IP address. Install your root filesystem image in this directory as root on the server, and export the directory tree with an entry in /etc/exports on the server, like:
/tftpboot (rw,no_root_squash)
If your system has a hard disk, you can start by using NFS then build a root file system on the disk and boot from that.
If your system has no ethernet, you may want to start developing on a board that does.
To make a diskless system standalone, you need an initial ramdisk image containing an ext2 filesystem to put in arch/ppc/mbxboot/ramdisk.image.gz. Then, build with "make zImage.initrd" and the ramdisk image will be mounted as the root filesystem at startup. See Documentation/initrd.txt in the kernel source tree.
You need to select both CONFIG_BLK_DEV_RAM and CONFIG_BLK_DEV_INITRD to build zImage.initrd. You also need a file in arch/ppc/mbxboot called ramdisk.image.gz. When you build zImage.initrd, the secondary boot loader is re-compiled with INITRD_OFFSET and INITRD_SIZE set, which are used to locate the start and end of the ramdisk.image.gz file in memory. The start and end are passed to the kernel in registers (r4/r5??), which it saves into the variables initrd_start and initrd_end. The secondary boot loader also changes the kernel command line arguments so that root=/dev/ram instead of root=/dev/nfs. The kernel does various things if initrd_start is non-zero, but the main one is to decompress the ramdisk.image.gz data into ramdisk 0, and because root=/dev/ram, this is then mounted as the root filesystem.
If your ramdisk is larger than 4 Mb, you will need to add ramdisk=xxxx to the kernel command line at boot time, or modify drivers/block/rd.c.
Beware that the CPU6 workarounds in the MontaVista 2.2.x kernel clobber the kernel command line, and cause the initial ramdisk mount to fail. See the thread at: http://lists.linuxppc.org/listarcs/linuxppc-embedded/200004/msg00103.html
There are a number of ways to create your inital ramdisk image, described below. For more info on building a root filesystem, see the Bootdisk HOWTO at: http://linuxdoc.org/HOWTO/Bootdisk-HOWTO-4.html
An example ramdisk.image.gz is already included in the Hard Hat kit.
You can also create your own on your development machine in a filesystem on /dev/ram. If your ramdisk is larger than 4 Mb, you will need to increase the default ramdisk size on your development machine accordingly.
LILO users can do this by adding the following line to the first section of /etc/lilo.conf:
ramdisk=65536
There is no real harm in asking for an excessive size, as /dev/ram* only allocates pages it actually needs to the ramdisk. However, you should use the "blocks-count" parameter to limit the filesystem size when you run mke2fs to prevent it creating unnecessarily large filesystem structures.
Another approach is to use the loop device on your Linux development host to mount the ramdisk image as a local filesystem, and then copy the files you require into it. To allow users to mount the ramdisk.image on /mnt/loop with "mount /mnt/loop", add this entry to your /etc/fstab:
/path/to/ramdisk.image /mnt/loop auto user,noauto,rw,loop 0 0
Note that the minix file system code in Linux is not endian-independant, so you can't build a minix file system image on an x86 machine and expect to read it on a PowerPC machine. ext2 does not suffer from this problem.
For more info, see the Loopback-Root-FS HOWTO at: http://linuxdoc.org/HOWTO/mini/Loopback-Root-FS-HOWTO.html
Search for ROMFS.
The 2.4 kernel series has a compressed read-only filesystem (cramfs) aimed at embedded systems, which can be back-ported to 2.2 kernels. If you're cross- developing, you need to modify
mkcramfsto swap between little and big endian.
ramfs from the 2.4 kernel is a simple filesystem ideal for use in a ramdisk. It can be used in combination with a cramfs read-only root filesystem, to mount writable filesystems on
/tmpand
/var, which typically need to be writable. This combination is ideal for systems which don't require persistent storage.
http://www.developer.axis.com/software/jffs/
JFFS allows persistent storage, optimised for flash memories rather than block devices like hard disks. It is aimed at providing a crash/powerdown-safe filesystem for disk-less embedded devices and is a better option than the cramfs/ramfs combination if your application requires persistent storage. You use it with the Memory Technology Device subsystem.