Re: Booting from gpt-on-raid1-on-gpt?

[Hauke Fath <hauke%Espresso.Rhein-Neckar.DE@localhost>]

On Sat, 28 Sep 2013 11:20:53 +0000 (UTC), Michael van Elst wrote:
> hf%spg.tu-darmstadt.de@localhost (Hauke Fath) writes:
> 
>> since I have just run into the problem again: Is there any perspective 
>> for booting NetBSD from a gpt-on-RAID1-on-gpt setup (see PR 44982) 
>> without falling back to disklabel(8)?
> 
> What part of the boot process fails?

Looking into things, I found that gpt(8) looks for a partition type 
that it considers bootable. And 'raid' is apparently not such a type, 
so gpt(8) will not install a mbr. The error message could probably be 
worded more clearly - I only understood what was going on after 
hexdump(1)ing the first block of the raw disk, and finding it empty.
 
> Booting requires
> 
> - MBR to fetch PBR

... see above - MBR not installed won't fetch PBR. I still have to find 
out whether a type 'raid' partition will have a PBR, and where that 
would be located - at the beginning of the 'raid' (raidframe) 
partition, or at the beginning of the root partition on the raid?

> - PBR interpreting GPT
> - PBR detecting RAID(?) volume and raidframe label.

My understanding is that PBR currently dowsn't do that. But my 
understanding is hazy, given the maze of little asm- and makefiles, all 
alike, that make up arch/i386/stand.

> - PBR interpreting GPT inside the RAID volume
> - PBR reading /boot
> 
> - /boot repeating about the same to read /netbsd
> 
> - /netbsd repeating about the same to mount /
> 
> Step 2 and 3 of PBR (same code as in /boot) probably
> need to be implemented.

Sounds spot on. I am still trying to tie the arch/i386/stand files to 
the steps you listed above...

Thanks for your comments.
hauke

-- 
Hauke Fath                        <hauke%Espresso.Rhein-Neckar.DE@localhost>
Ernst-Ludwig-Straße 15
64625 Bensheim
Germany