Re: CVS commit: src/distrib/utils/sysinst

[David Laight <david%l8s.co.uk@localhost>]

On Fri, Jul 16, 2004 at 02:32:56PM -0700, Bill Studenmund wrote:
> On Thu, Jul 15, 2004 at 09:47:36PM +0100, David Laight wrote:
> > On Mon, Jul 12, 2004 at 10:33:21PM +0000, Christos Zoulas wrote:
> > > 
> > > Log Message:
> > > PR/26201: Jukka Salmi: sysinst dumps core on NetBSD 2.0_BETA i386
> > > The problem is that current_cylsize is used in the menu to select
> > > rounding, and it is un-initialized at this point.
> > > XXX: Must be pulled up to 2.0
> > 
> > Needs to be fixed correctly first...
> > 
> > Rightly or wrongly, partitions have always been aligned to BIOS cylinders
> > on i386 - the committed change (unless I've misread it) uses the
> > 'real' (aka faked by the disk) cylinder size instead.
> 
> Does "Needs to be fixed correctly" really go with "Rightly or wrongly"? 
> :-)
> 
> While I think that a change like this perhaps should be documented, it 
> doesn't sound like a bad change.

I've been looking at this code for a couple of days (on and off) trying
to decide what it does, and what it should do.

All the 'rounding to cylinder boundaries' for the netbsd label does
seem to be done with the disks geometry (not the bios one), although
user input would be in bios cylinders - if you edited the mbr.

I was thinking that the code needed to set the bios cylinder size
(into current_cylsize) before embarking on the disklabel setup, but
it is probably better to just reset to input in MB.

Indeed even for sparc (where the sun label requires partitions be aligned
to cylinder boundaries) there is no need to enforce input in cylinders
as values input in MB are always adjusted to align to cylinder boundaries.

I might remove the md_set_sizemultname() completely, and the call to
ask_sizemult() before the mbr edit.

Further enhancements would be input in GB, and the ability to select the
partition alignment (for non-sparc).

        David

-- 
David Laight: david%l8s.co.uk@localhost