Matthias Drochner <M.Drochner@fz-juelich.de> writes:
> Besides this - I have the feeling that using "real" pointers
> instead of indices into the locator array within the cfdata
> would be a good thing - especially for loadable drivers.
> Didn't look at the issue for a while however, so I can't back
> it with hard arguments...

Right, so, once you start gutting stuff here:

* almost easy: change the locator type to intptr_t (do we have that
yet?) or whatever it is, and provide the ability to use string-valued
locators.

* not so easy: figure out what the right thing to do for locator
tables really is.  possibilities:

	* completely independent arrays per device?

	* pointers into single large array?

	* (table, index) pairs?

	* "other"?

* worse: figure out what the right thing to do for parent vector
tables really is.  You want to be able to glue a loaded device into an
existing device tree, when loading, which means teaching the kernel
how to resolve parent vector references.

(I think your long-ago PR on the subject is probably one step in the
right direction, but it doesn't go nearly far enough if i recall it
correctly, and it was never clear to me that doing it only partially
was the right thing.)



cgd
-- 
Chris Demetriou - cgd@netbsd.org - http://www.netbsd.org/People/Pages/cgd.html
Disclaimer: Not speaking for NetBSD, just expressing my own opinion.