I wasn't missing this- this is fine insofar as it goes. You can use the
kernel as it moves the place where one defines this platform's
char/block major+minor set from a conf.c to a config file.

I don't have an objection to what is being proposed. I wanted to
understand if the proposers were clear that the last steps were
missing. This is possibly a step in the right direction.

On Thu, 28 Jun 2001, Robert Elz wrote:

>     Date:        Thu, 28 Jun 2001 12:50:09 +0200
>     From:        Ignatios Souvatzis <ignatios@theory.cs.uni-bonn.de>
>     Message-ID:  <20010628125009.A28656@theory.cs.uni-bonn.de>
>
>
>   | Well, but as long as the userland problem isn't solved, you can't use your
>   | changed kernel at all...
>
> I think that you and mjacob were both missing ...
>
>   4.2.	Userland - config(8)
>
>   4.2.1.	Grammer
>
>   device-switch <name> chr <num> [blk <num>] [<options>]
>
>   name		- The prefix of bdevsw/cdevsw entry (required)
>   chr		- A character major number (required)
>   blk		- A block major number (optional)
>   options	- Conditions to determine whether should be attached or not
> 		  (optional)
>
> That is, the major device numbers are still being wired into the
> kernel, they're not being dynamically allocated - just now that info
> is coming from the config machinery, rather that someone hand editing
> the conf.c file and corectly positioning the list of functions in the
> [cb]devsw tables.
>
> Really, if Matt Thomas looks at this and says "real cool" or something,
> it really can't be a totally broken design, can it?

As much as I respect Matt, the answer to that last is "can't tell". I
don't know if Matt Thomas has ever worked on the issues inside kernels
that come from this topic. I have.

-matt