> > Why does the PCI spec care whether the code that mucks with these
> > registers is linked into the lump called firmware or the lump called
> > the OS?
> 
> Because the firmware knows more about the machine state than any OS ever
> could, unless you have a few million and vendor support.

But why is the PCI spec the place for this decision?

The only vendor support needed should be proper documentation.
(Yes, I know, many vendor don't want to provide docs.)

> I take the other approach -- I'd like hardware _drivers_ to be in the
> firmware, and called from NetBSD.  Like open firmware, sorta.

Which leads to things like having an x86 emulator in the Alpha SRM.  :-(

And unfixable bugs.  :-(

And unfixable interfaces.  (ARC, anyone?)

-Dave