Subject: Re: FASTTRAK100 woes (Was: 3ware escalade install problem)
To: None <tech-kern@netbsd.org>
From: Manuel Bouyer <bouyer@antioche.eu.org>
List: tech-kern
Date: 05/20/2003 22:47:08
On Tue, May 20, 2003 at 05:53:45AM +0200, Lucio De Re wrote:
> I will believe that!  Even if it's not quite the way it was presented
> to me.
> 
> Does that mean that (a) I have to learn about RAIDframe to match
> the features of the controller and (b) as David Laight suggests,
> there's no way to boot from the actual RAID device (reliably?  It
> does get some of the way, seemingly).

(a): yes
(b): i'm not sure what he means. But if you can configure the BIOS to just
     have 2 simple disks instead of one raid, there's a possiblility.
     See below.

> 
> Alternatively, do I just scrap the idea altogether until I can
> afford equipment to test on?  I'll probably have to do that anyway,
> because there isn't enough room in the chassis for a boot drive,
> unless I resort to PXE and/or diskless booting (if I ever figure
> that bit out :-)

If you can setup the BIOS to use (and present to booted software) 2 disks,
you can then setup a raidframe and boot from it.

The way of doing it is to have a small FFS partition on both drives for the
/boot. If the 'a' partition is marked type RAID, and it's a raid-1, it knows
how to load a kernel from here.
The raid which has the root paritition has to be on wd0a (and wd1a).
The partition with /boot has to be at the start of the NetBSD partition.
I just setup a system like this, here is the partitionning I used:
8 partitions:
#        size    offset     fstype  [fsize bsize cpg/sgs]
 a:   2207520      1008       RAID                      # (Cyl.    1 - 2190)
 c: 240121665        63     unused      0     0         # (Cyl.    0*- 238215)
 d: 240121728         0     unused      0     0         # (Cyl.    0 - 238215)
 e: 237913200   2208528       RAID                      # (Cyl. 2191 - 238215)
 f:       945        63     4.2BSD   1024  8192   103   # (Cyl.    0*- 0)

(a and e both hold a raid1, the first one for / and swap, the second for
everything else).

Of course you can't do this from sysinst. The way I do it is to install
on the first disk, and use this system to setup the raids on the second
disk and make the disk bootable, using a nonexistant drive for the
mirror part of the raid. raidframe will mark the second component
as failed.
Once disk1 is bootable, with the raid parts, I boot from disk1, 
redisklabel disk0 to match disk1, add disk0 partitions as hot-spares and
start reconstruct of failed components.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 24 ans d'experience feront toujours la difference
--