On 1/24/21 11:52 PM, Brook Milligan wrote:
The drive I am trying to replicate manually is from an evbmips (octeon) system that is working fine and was (more or less) created by NetBSD tools for a release.
This is key information. I'm guessing, in the past, that the hpcmips boot loader (u-boot) didn't recognize the GPT MSDOS partition that contained bootefi. I wonder if that's still the case.
You could try deleting the the msdos partition from the MBR (I'd back the MBR up first) and see if the machine still boots. If so then you should make the GPT protective partition in the MBR cover the entire disk.
If the MBR MSDOS partition is needed then you should have the GPT protective partition start at 196608 (32768 + 163840) and cover the remainder of the disk. Are you looking to script the creation of the disk image? If not, sysinst should create the GPT partitions with no difficulty. You'll need to add the MBR partition(s) by hand.
If you're having getting the machine to boot, you may be missing the u-boot code that's in the blank space before the start of the MSDOS partition (sectors 34 to 32767). You could copy the u-boot code using dd or look for in pkgsrc for a package that will build the boot code for you. Such a package would give you the command to copy the boot loader to other disks.
Having the GPT partition table start at sector 1 shouldn't be a problem. An unscientific survey of the machines I have with GPT tables shows that all of them have the GPT header in sector 1.
HTH, Jason M.