From: Adrian M. <ad...@mc...> - 2002-06-06 21:56:05
Attachments:
nfs_boot.html
|
A long time ago (about 8 weeks) I promised on the IRC channel that I would write an article for the web site on NFS booting. Here is the first - very rough draft. I have discovered technical documentation is not for me. Please let me have your comments (be as nasty as you like) - especially on the tech which is a bit vague at times! Adrian |
From: Paul M. <le...@Ch...> - 2002-06-07 18:33:33
|
Looks good. Some cosmetic fixes listed below.. On Thu, Jun 06, 2002 at 10:57:47PM +0100, Adrian McMenamin wrote: > completely and bott your kernel straight off the network ^^^^ s/bott/boot/ > 10MBit port and have heard they are a bit flakely at higher speeds - s/flakely/flakey/ > but cannotconfirm that from personal experience. ^ s/\(cannot\)\(confirm\)/\1 \2/ > BBA's are not cheap - not mny were made and demands is high. Mine cost s/mny/many/, s/\(demand\)s/\1/ > Of course other tools are available - but I recommned dcload-ip: I > know it works and its simple. > =20 > (NB: If you upgrade the kernel on your Linux box you will have to > recompile the server (host) end of dcload-ip - but the Dreamcast end > doesn't have to change). Neither myself or Marcus have managed to get dcload-ip to work sanely out-of-the-box. I've got a patch up here: http://stampede.org/~lethal/sh/dcload-ip-1.0.3-suck-less.diff that fixes the issues. Haven't tried newer versions. > 3. Compile your DC kernel to support NFS booting > =20 > Building a kernel is covered elsewhere on this site. But you have to > compile a version that supports NFS booting for this to work. The key > points to remember are: > =20 > * TCP/IP networking (obviously) must be on > =20 > * EISA, VLB, PCI and on board controllers should be selected > + RealTek RTL-8139 PCI Fast Ethernet Adapter support should be > selected, along with "Support for Dreamcast Broad Band > Adaptor" > * Then, under file systems support make sure you select: NFS file > system support, with root file system on NFS, and you should also > select support for version 3 client support (assuming you are not > mounting off some very ancient box) Due to stupidity in the kernel, you also have to have IP autoconfig turned = on in order to see the root on NFS option. Thus, you need the following in your =2Econfig: CONFIG_IP_PNP=3Dy CONFIG_NFS_FS=3Dy CONFIG_ROOT_NFS=3Dy optionally, you can CONFIG_NFS_V3=3Dy if you want NFSv3 support. > * I am also using automounter support, but I have no idea if that is > necessary - perhaps someone could tell me! > =20 No, automount is not necessary at all. You setup the automounter to automagically exec a mount command when you traverse a mountpoint. It's cut= e, but hardly useful, especially on the DC. > To get NFS to bbot properly, I had to tweak it just a tiny bit (I was s/bbot/boot/ > I have also used the Busybox sheel as my initrd, and bound that with s/sheel/shell/ > 6. Pipe it accross and watch it boot. > =20 s/accross/across/ Regards, --=20 Paul Mundt <le...@ch...> |
From: Adrian M. <Ad...@mc...> - 2003-04-06 16:00:31
|
I have finally got my NFS daemon working again - and fixed the other configuration problems in my ia32 kernel. I am specifying nfs as the root filesystem on the DC but it is just loading the ramdisk - should I be passing over and empty initrd for this to work? Adrian |
From: Karl T. K. <ka...@li...> - 2003-04-12 13:07:18
|
On Sun, Apr 06, 2003 at 04:58:52PM +0100, Adrian McMenamin wrote: > I have finally got my NFS daemon working again - and fixed the other > configuration problems in my ia32 kernel. I am specifying nfs as the > root filesystem on the DC but it is just loading the ramdisk - should I > be passing over and empty initrd for this to work? I can't remember that being necessary. The kernel will use the nfs device as root (/dev/nfs) and look for /bin/init on it (or whatever you init= setting is). The caveats being you having NFS support in the kernel, and the IPs configured correctly. Use tcpdump or other diagnostics (nfs log) to see if you actually have a request from the DC at boot. Kind regards, Karl T |