Re: Path to kmods

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

On Thu, Nov 13, 2008 at 12:05:24PM +0000, Andrew Doran wrote:
> On Thu, Nov 13, 2008 at 12:00:33PM +0100, Manuel Bouyer wrote:
> 
> > Sure. But this doesn't mean that the types, inline functions, etc ... used
> > for native and PV kernels are the same (this is the difference between
> > API and ABI). I even think it'd be hard to have them the same, unless we
> > define them to be opaque pointers. But I guess this will add a mesureable
> > overhead to critical paths.
> 
> Your argument is short on detail. You mentioned PAE, but what other issues
> do you see causing incompatibility?

Lots of things which are in machine-dependant headers (different
#define, different inline functions, different structures). Some of them are
part of the kernel API. If you want a specific example, look at pcitag_t.
Or PCI_MACHDEP_ENUMERATE_BUS. or pci_intr_handle_t. 

Not to mention things like struct cpu_info which are of different size,
and may be used by modules though the maze of includes and inline functions.


> Which of those issues are intractable?
> Have you tried loading an i386 kernel module on a xen kernel?

No need. I'm sure a a device driver module won't work.

-- 
Manuel Bouyer, LIP6, Universite Paris VI.           
Manuel.Bouyer%lip6.fr@localhost
     NetBSD: 26 ans d'experience feront toujours la difference
--