On Wed, Aug 25, 2004 at 10:03:20AM -0400, Jan Harkes wrote:
> I am considering that it might be better to just get userspace to work
> correctly when compiled as 64-bit (i.e. aim for 64/64 instead of 32/64).

Hello Jan,

I think it would be a very good move to make the kernel-user interface
independent of which mode either of the peers is in.

Think that some systems can be temporarily rebooted into another mode,
(32 vs 64 bit or as an extreme example even bigendian vs littlendian)
and usually it does not imply replacing/maintaining in parallel all of
the userspace, does it?

A well-defined interface would work the same.
I would not expect a remarkable performance penalty caused by {un,}packing
the data passed.

> Well, I am always hesitant to break the kernel interface, so that part
> might go in a bit later. I'll try to get most of the stuff that doesn't
> force a kernel protocol change in the next release.

It looks like the interface is in need of some incompatible changes,
then let's do it in a clean way - so that it does not put size limitations
on the data being passed, and that we define it independently of the
"default word size" or whatever, but rather in absolute terms.

the usual 2c ...
--
Ivan