|
 Home Page
|
|
|
Documentation top level
|
If you are unfamiliar with NetBSD or how to layout portions of the system, take a look at The NetBSD Guide.
Since the kernel does not have the ramdisk image to support an installer it has to have its root filesystem provided externally. The NFS server must be able to serve out not just files, but also device nodes (/dev entries). This may not be possible with non-Unix-like NFS servers. First one needs to prepare the server and then netboot the kernel. See the NetBooting question for details. After setting up the diskless root, you will want to copy the sets over for easy installation.
The original disk in the cobalts came partitioned for Linux. The first FDISK partition is special (Linux root (/)) and must always exist. This is because the firmware looks to there for the kernel to boot. The remaining partitions can be removed and the space used for NetBSD. See the preparing the harddisk for installing NetBSD question for instructions.
Note: If your system did not have Linux on it before, tools to build an ext2fs in the first FDISK partition can be found in pkgsrc under pkgsrc/sysutils/ext2utils
The last step is to configure the system. The NetBSD Guide can assist you in this.
Press the space bar while the diagnostics run to halt the autoboot process.
Type ? at the firmware prompt to get a list of commands.
bfd /netbsd.gz nfsroot=/home/raq/root
The firmware is picky about syntax and in general, so if things fail mysteriously, try to conform to the conventions described above. For netbooting, you need to NFS-export the directory given to "nfsroot=", and the named kernel ("/netbsd.gz") needs to be in that directory. You will also need to setup rarp and dhcp. Once the kernel is loaded with the command line values, the data given via DHCP is used to mount the root filesystem. Here is a DHCP entry that is known working:
host raq {
hardware ethernet 0:10:e0:0:52:62; # raq MAC
fixed-address 10.0.0.15; # raq address
filename "/netbsd.gz"; # kernel name in root-path
option root-path "/home/raq/root"; # absolute dir on nfs server
server-name="10.0.0.3"; # IP of nfs server
}
See
the Diskless NetBSD HOW-TO.
Also, various models expect different filenames for the kernel as noted above. You can use the /usr/mdec/installboot script to create them all.
The default autoboot process loads a gzip'ed ELF kernel from /boot/vmlinux.gz (or /boot/vmlinux_RAQ.gz or /boot/vmlinux_raq-2800.gz) from the EXT2 filesystem on the first MBR partition on the primary disk.
There is no provision for changing the boot path beyond disabling autoboot and the kernel (or the boot loader, if/once we should write one) must always be gzip'ed.
When there is no NetBSD disklabel on the disk, the four MBR partitions are mapped to [e-h].
Everything else on the Linux filesystem can be trashed as it is totally unnecessary. You should be able to just fdisk/partition/newfs and copy a kernel onto the new filesystem. You shouldn't need to do a complete install on the disk.
The disk that comes in the Raq2 has four partitions:
All the Linux partitions are of ext2 filesystem, and they can be checked (if needed) and mounted from NetBSD with:
fsck -t ext2fs /dev/wd0e fsck -t ext2fs /dev/wd0g fsck -t ext2fs /dev/wd0h mount -t ext2fs /dev/wd0e /mnt mount -t ext2fs /dev/wd0g /mnt/var mount -t ext2fs /dev/wd0h /mnt/homeBefore installing NetBSD, it is recommended to back up the data of the Linux /home partition to the Linux /var partition. Then, use NetBSD's "fdisk -u" comman to change the partition ID from 131 (Linux userland) to 169 (NetBSD). Following a reboot to get the new partition data, a disklabel can be written:
disklabel wd0 >x disklabel -r -R wd0 xThe disklabel now lacks a NetBSD root partition (wd0a), and we use the old Linux home partition (wd0h) for that. Just change the "h:" in "disklabel -e wd0" to "a:" and you're set. Next, create a filesystem and mount it:
newfs /dev/wd0a mount /dev/wd0a /mntAfter that, the NetBSD sets can be extracted from NFS. You need to extract at least etc, base and kern, comp is strongly recommended, the others (text, games, man, xbase, ...) don't hurt. The sets are in the "cobalt/binary/sets" of the latest NetBSD release:
tar tar pzvxCf /mnt etc.tgz tar tar pzvxCf /mnt base.tgz tar tar pzvxCf /mnt kern.tgzAfter that, the "/dev" dir needs to be populated:
cd /mnt/dev sh MAKEDEV allThe final step is to create a bootable kernel in the right place. For that, we compress the kernel from the "kern" set, and put it on the Linux root dir. Setting up an entry in (/mnt)/etc/fstab to do so is recommended:
/dev/wd0e /altroot ext2fs rw 0 1With that line in (/mnt)/etc/fstab, we can mount the Linux root partition, and drop our kernel in the expected place:
mount /altroot cd /altroot/boot mv vmunix.gz vmunix.gz.linux cp /netbsd vmunix gzip -v9 vmunixNote that the kernel needs to be presend (and in sync!) in the following places:
After these steps, the firmware should be able to find the kernel and boot it:
rebootWhat follows is that the system will boot and needs to get configured. If you have trouble mounting the root filesystem read/write, the command to get that is "mount -u /dev/wd0a /". Now setup /etc/rc.conf and enjoy!
Sorry, there's no all-singing, all-dancing install-routine (yet).
This serial interface is however not a full I/O serial due to both hardware and software limitations. It can only be used while talking with the boot ROM or reading console about which can come in handy while debugging.
In your Qube2700 you should find a print on the PCB as (or similar):
Cobalt Microserver, Inc.
CPU Board 850-00101
You would then have a 18 or 20 pin header titled "SCC CON" adjacent to the AMD 28F040 chip. The pinout of the header is as follows:
memAD0 2 1 +5V
memAD1 4 3 Dadr0
memAD2 6 5 Dadr1
memAD3 8 7 serialInt_low
memAD4 10 9 serialClk
memAD5 12 11 serialCs_low
memAD6 14 13 tied to devWr_low
memAD7 16 15 no-connect
devRd_low 18 17 ground
devWr_low 20 19 ground
The pin 20 and pin13 tie is because there are some systems with 18-pin headers, and some with 20-pin.
The schematic is made with OrCAD and requires only an 85c30 serial chip, a DS14C239, and a few resistors and caps to build.
5.Bank 0:.................................16M 6.Bank 1:.................................0M 7.Bank 2:.................................16M 8.Bank 3:...............................**Simm0 must be >= Simm1**The above example is for a 16MB module in the slot near the case, and a 32MB module in the other slot. With such a config, you can experience all sort of weird problems. If you exchange them, all will be good. The boot message should be:
5.Bank 0:.................................16M 6.Bank 1:.................................16M 7.Bank 2:.................................16M 8.Bank 3:.................................0M
# Determine machine name. Could as well hardcore different
# install-kernel-machinename install targets for different machines.
MACHINE_NAME!= uname -n
# Place where the ext2fs partition is mounted, there should be a
# "boot" subdir where the kernel "vmlinux.gz" is located
BOOTDIR?= /altroot/boot
# Build name for new kernel - can do something more exciting with
# using date etc., override by setting in /etc/mk.conf.
.if !defined(NEWKERNEL)
NEWKERNEL!= echo vmlinux-`sed -n < vers.c '/osrelease/s/.*"\(.*\)".*/\1/p'`-`date +'%Y%m%d'`-`cat version`.gz
.endif
install-kernel-${MACHINE_NAME}:
gzip -c9 netbsd >${BOOTDIR}/${NEWKERNEL}
rm -f ${BOOTDIR}/ovmlinux.gz
ln ${BOOTDIR}/vmlinux.gz ${BOOTDIR}/ovmlinux.gz
ln -f ${BOOTDIR}/${NEWKERNEL} ${BOOTDIR}/vmlinux.gz
rm -f /netbsd
cp netbsd /netbsd # for vmstat and friends only
With that, 'make install' should put the kernel into the right
location(s).
|
|