[ipxe-devel] DHCP failing on Intel 82579V Gigabit [8086:1503] after PXE boot]

[Michael Brown]

On 05/03/13 15:31, Robin Smidsrød wrote:
>> In the longer term, I think the optimal solution would be to fix up
>> ipxe.kpxe (which isn't currently a sensible build target) so that it can
>> use a native driver if present, falling back to the UNDI driver if
>> necessary.  This would allow native iPXE drivers to be used instead of
>> the UNDI driver where possible, which should obviate the need for this
>> hack.
>
> I would very much like to know what is missing for this thing you
> suggest here to actually work. Because it does seem to be an even better
> "default" chainloader than undionly.kpxe currently is, as it would
> include fallback if native drivers aren't available.

We need a sensible plan for conditionally unloading the PXE stack if and 
only if the underlying UNDI driver corresponds to a device for which we 
have a native driver.  This doesn't fit neatly into the current driver 
model, since PCI and "undionly" are treated as completely independent buses.

We would probably want to start out by attaching the underlying UNDI 
driver as an UNDI device, and provide some mechanism to allow 
drivers/bus/pci.c to detach any drivers currently attached to the same 
physical device whenever it finds a native driver.  I haven't yet 
thought of an elegant way to do this, or about the potential complications.

Michael