On Mon, Feb 20, 2006 at 07:34:14AM -0600, Peter Seebach wrote:
> The conceptually correct fix is that, if a SPECIFIC CONFIGURATION needs to
> attach devices in a given order, that's what should specify the order, and
> the code to handle it should be generic.

Sure, and we should have a "do what I want" compiler option.

What we can do in code, right now, is to order pci devices, use 
config_interrupts() or config_defer(). On direct configuration buses
I think config(1) can do none of those, and I see no easy way to teach
it any of them.

I see other requirements for "guiding" autoconfig order, like walking the
ACPI device tree and constructing an autoconfig tree during the walk - but
that is something completley different.

In your case to me it looks a lot like a bad hack to abuse autoconfig for
all of this. Maybe you should describe the problem on tech-kern and ask
for options.

Martin