Adam Agnew <[EMAIL PROTECTED]> writes: > > > > > Could you give me more detail on your current status > > > about LinuxBIOS + IDE boot ?? I am looking for any solution > > > to get rid of DoC. > > > > That's interesting, here at the Lab we are looking for any solution to get > > rid of disks (on cluster nodes, I mean). > > > > Then if I, or some other kind kind patient sole, wanted to maintain a > bootloader tree to its upmost usefullness, would you rather have two > seperate bootloaders for each task, or one monolithic (but still modest) > bootloader? Input is really welcome.
I want a bootloader that I can stuff with linuxBIOS in 64KiB. This makes fallback bootloaders much easier to manage. My gut reaction is that I want one monothlithic bootloader. But size is pretty important. > What I've done Ollie is to take the ext2fs code and polled ide code from > RedBoot. I then made a quick kludge where by the new etherboot reads a > Multiboot elf image (just like eric's mkelfimage makes) named /kernel from > hda1 and then passes it on to etherboot's elf loader as if it was just > fetched off of the network. Better, more mature code would prompt you for > the path and command line and no doubt this will be better and more mature > as soon as i get a lot of time. > > There are two minor problems with this attack. > > 1) All the RedBoot code needs to be rewritten or replaced to conform to > the redboot license. This is stupid because they, redhat's people that is, > are getting it changed over to a dual lgpl license shortly anyway. But, I > requested for them to look the other way for a couple of weeks while I > sent this code out for more eyes to look at, but they never responded. > Anyway, I wanted a proof of concept most of all and I have one. > > 2) Somewhere in that ext2 code or in the polled ide code is a bug. This > bug was present when it was in RedBoot as well. I'm hoping the bug is in > the ext2fs code and it appears to be. I'll be replacing that ext2 fs code > anyway to stay with all gpl code. I'll probably get it from grub as they > have a wealth of file systems ready to roll. So, it can load /kernel on > several different hard drives, but not from others. I imagine that > this is some kind of classic you can't read over 1024 cylinders kind of > problem. I haven't investigated it closely yet. > > Altogether, this is ext2fs code, polled ide code, then everything that > comes with etherboot when you compile for the sis900. It weighs in at > around 24k this way. Not too shaby.. Nope. I do have a comment though KISS. For complicated super featured things we can build the all doing all powerful bootloader on top of the linux kerenl. Able to read an write all filesystems, and many other things. For the fallback and simple cases we want to be simple. If you support a filesystem support just one filesystem. Loading an ELF image out of a partition with a special type would be sufficient for me. Or always finding your ELF header on the first or second sector of the disk, and simply not caring about partition tables at all. Simple and stupid is perfectly fine. That is why I picked the ELF format. It is simple and stupid but it has just enough power to be convinient as well. All the ELF program says is load this chunk of file into this chunk of RAM. Eric
