Partitioning proposal for boot loader
Now that the prelimary bootstrap test appears to work, I am preparing to do the necessary work for actually loading the (future) kernel image.
As I mentioned in previous mailings, the bootloader needs to understand how to find the kernel on the disk.
I propose that for the near-term, we use a simple PC-like partition scheme for HPPA/Linux. In the future, once Linux is up and running,
we can work on co-existing with HP-UX on the same physical disk (see below for more on this). This will keep the bootloader very small and simple, and allow us to quickly move to working on the hard part, the actual kernel.
My idea is to create simple partition table structure which would contain (for each partition) the starting sector, length, and perhaps a flag word. This table would be of fixed length (say 16 entries) and be stored at a known place near the beginning of the disk.
The kernel itself could be stored in a non-filesystem partition (i.e. just a raw partition) so the bootloader only has to read in the data and not have need a standalone filesystem.
Unless there are strong objections, I will proceed with the coding of this scheme over the weekend.
Regards, Jason.
----------------------------------------------------------------------------
HP partition information (fyi):
After investigating my 10.20 drive, and some HPUX header files, I think I understand (at least partially) how the partition scheme works. Unfortunately, it doesn't appear to me that their scheme allows another OS to coexist peacefully on the same disk.
Basically, they store a partial partition table in the LIF directory of the disk in a LIF file called LABEL. The table only contains information for 4 partitions -- BOOT, ROOT, SWAP, DUMP. This is enough for the hpux loader to find the kernel image and boot. It seems that the LVM partition information is internal to HPUX itself. LVM only seems to 'export' the info for the 4 partitions mentioned above into the LABEL file, even if LVM has setup more partitions.
What it amounts to, is that it would be easy for the bootloader to read the LIF directory, pick up file LABEL, and get the partition information. But if we write our own entry into the table, we would clobber some other LVM partitions since they aren't reflected in the table.
Its a little strange. If someone from HP (or someone who ACTUALLY knows this stuff) thinks I missed whats happening, please let me know. This is what I could get from looking at hex/ascii dumps of the lif volume and a few header files...
----------------------------------------------------------------------
To unsubscribe: send e-mail to [email protected] with `unsubscribe' as the subject.
Quoted 7 lines · Jason Eckhardt
As I mentioned in previous mailings, the bootloader needs to understand how to find the kernel on the disk.
I propose that for the near-term, we use a simple PC-like partition scheme for HPPA/Linux. In the future, once Linux is up and running,
we can work on co-existing with HP-UX on the same physical disk (see below for more on this). This will keep the bootloader very small and simple, and allow us to quickly move to working on the hard part, the actual kernel.
I think your order of doing first pc-like partitioning then hpux partitioning is correct, but I think that partitioning itself isn't needed right away.
To start, I'm really much keener on net booting. This is because the kernels will be cross compiled on other machines, and it lets us get straight to the testing, and we don't need any disk reading or anything like that.
But certainly sorting these partitioning issues out will be important down the line.
Quoted 4 lines
My idea is to create simple partition table structure which would contain (for each partition) the starting sector, length, and perhaps a flag word. This table would be of fixed length (say 16 entries) and be stored at a known place near the beginning of the disk.
I'd much rather use an existing paritioning structure if at all possible.
- Alex
----------------------------------------------------------------------
To unsubscribe: send e-mail to [email protected] with `unsubscribe' as the subject.
Quoted 6 lines
I propose that for the near-term, we use a simple PC-like partition scheme for HPPA/Linux. In the future, once Linux is up and running,
we can work on co-existing with HP-UX on the same physical disk (see below for more on this). This will keep the bootloader very small and simple, and allow us to quickly move to working on the hard part, the actual kernel.
My personal opinion is that they should go on seperate physical disks. If I install two versions of HPUX on a box, I use seperate disks.
If you want to put them on the same disk, this means you can't use LVM under HPUX, or that Linux has to support HP's LVM.
Rich
----------------------------------------------------------------------
To unsubscribe: send e-mail to [email protected] with `unsubscribe' as the subject.
I also agree that if we want different OS in a single box we should rely on seperate disks. HP-UX doesn't support different OS 'es on a single disk and doesn't really support partitions in general other than allocating a swap area (except LVM). BTW, there is a page on LVM for Linux. I can dig this up and this would be something good to incorperate in HPPA-linux since this is what is native for most HP-UX systems these days.
Signature · 9 lines
Steven A. DuChene Linux Fan! http://www.mindspring.com/~sduchene
Outside of a dog, a book is a man's best friend. Inside a dog it's
too dark to read.
-- Groucho Marx
----------------------------------------------------------------------
To unsubscribe: send e-mail to [email protected] with
`unsubscribe' as the subject.