Re: loongson2f: evbmips or new port ?

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Thu, Aug 11, 2011 at 11:28:36AM -0700, Matt Thomas wrote:
> 
> On Aug 11, 2011, at 11:21 AM, Manuel Bouyer wrote:
> 
> > On Thu, Aug 11, 2011 at 01:29:44PM -0400, Michael wrote:
> >>> compiled with -DUSE_MBR but evbmips is not there.
> >>> Note that USE_MBR only changes the default value of a flag, so
> >>> it may be possible to set it contidionally on a sysctl value.
> >> 
> >> Needs to be added I guess - I formatted an SD card on an amd64 box for use 
> >> with the gdium.
> >> So far I only worried about the kernel reading the disklabel which works 
> >> just fine.
> > 
> > I just tried, with switching my loongson to kern/subr_disk_mbr.c.
> > It mostly works, but it seeems that the disklabel and mbr
> > data structure overwrite each others: when I write a new disklabel
> > I loose the first 2 mbr partitions, and when I write a mbr with
> > fdisk, disklabel complains about invalid partition 'p' which I never
> > entered.
> > This is probably because of:
> > #define LABELSECTOR     0                       /* sector containing label 
> > */
> > #define LABELOFFSET     64                      /* offset of label in 
> > sector */
> > 
> > in evbmips's disklabel.h. LABELSECTOR may be OK but LABELOFFSET may needs to
> > be changed. Or maybe something has to be changed in disklabel(8),
> > I didn't look further yet.
> 
> I'd just change them to match i386.

If we do this, then existing evbmips installations won't find their
disklabel any more :(
In addition I think the issue is in disklabel(8) itself, not really with
LABELOFFSET/LABELSECTOR: with a mbr the disklabel is supposed to be at the
start of the NetBSD partititon, not in the mbr itself.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--