Re: "ide=reverse" do we still need this?

[Ken Moffat <zarniwhoop@ntlworld.com>]

On Wed, Feb 13, 2008 at 03:32:49PM +0000, Ken Moffat wrote:
> On Tue, Feb 12, 2008 at 08:43:04PM -0800, Greg KH wrote:
> > 
> > Can't you just boot with /dev/disk/by-id/ and an initramfs to not have
> > to worry about such a thing in the future?
> > 

 Initramfs isn't something I've ever tried, so I'm not about to rush
into it on the server.  Maybe I'll try it on a desktop one day.
Anyway, I've now got it running without ide=reverse, details follow
for anybody else who gets a similar problem in the future.

>  Can comebody remind me what the initramfs is for in that situation,
> please ?  From the little I've noticed, I thought the /dev/disk/by-id
> part went into fstab ?  At the moment, I just build the things I
> think I need in to the kernel on that box, without modules.
> 
 And for the next person asking this, it seems to be so that you
can specify the root= parameter.

>  Anyway, I'll try to find time to read my notes to see if I can
> identify what happened/when, and to take the box down again so I
> can try to confirm exactly what the problem is, if it still exists.
> I certainly won't be taking it down until I've written my weekly
> backups to tape at the weekend, so maybe not before next week.
> 

 Turns out I was wrong about having a SATA disk for the system - I
used to, but then I needed to separate the backups into separate r/o
and r/w filesystems when nfs no longer let me export part of a ro fs
as rw.  In the change, my staging area for writing to tape or DVD
moved to the system disk, and the only big-enough disk I could get
locally was parallel ide.  So in that setup, ide on a card comes
first.
> > Have you tried the PATA drivers instead of IDE to see if this solves the
> > "moves around" issue?  If they work, then you would not need the command
> > line option at all.
> 

 My first thought was to try using libata for the drives on the
add-on card (sii0680), although it's marked as experimental.  Maybe
I picked the wrong driver, but they didn't show up.  Reverted to
previous config.

 Changed to mount-by-label so that I don't have to change fstab for
the old and new kernels.

 Moved the main drive and the CD to libata (sii still old IDE) -
specify sda instead of hda in root=, change system scripts referencing
drives by name (for SMART - system disk is again -d ata, the data raid
moved from hd{e,g} to hd{a,c}), now seems to be working but I expect
I'll notice a few things more in my scripts over the next days.

 I also discovered that lilo needed the real node specified in root=
to exist.  It complained, so I added a symlink to the hdaX node -
panic'd trying to load rootfs from 0307.  Reboot to old kernel, run mknod on /dev,
repeat, booted.

Ken
-- 
das eine Mal als Tragödie, das andere Mal als Farce