Re: CVS commit: src/sys/kern

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

> but unfortunately all of them don't have the same API.

This isn't an API issue!

> On i386 and arc MBR_PART is 3, but it is 2 on sh3, arm and prep.
I hope you mean RAW_PART....

> arm ports assume that disklabel is top of RAW_PART,

> but i386 (and some ports) assume it is top of C.
Actually the get the disklabel from the start of the first netbsd
MBR partition, then make (or hope) that it is partition C.

> The other ports assume it is top of A.
There is a further constraint on some ports (that code might rely on) that
the root filesystem (or more particularly the boot filesystem) has to start
of at the beginning of the disk.

> (then bounds_check_with_label() won't work on some ports BTW)
RTFC, it would find the disklabel in all the above cases.
(Although not necesarily write it back).

> There is no unified API yet even for MBR, so we should define it first.

Making other ports use that code will (and I knew it would) require
some tweaks.  But the intention is to use MI code.

> IMHO, RAW_PART should be 2 on all ports.

I'd go for making it 255 (actually stealing enough of the 'device' part
of the minor number to get one minor per device so that the 'entire
disk' device number doesn't (have to) correspond to a partition.
This is probably necessary in order to be able to read disks on (netbsd)
ports other than the one on which they were written.

        David

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