On Sun, 23 Jul 2000, Brett Lymn wrote:

> According to Matthew Jacob:
> >
> >This is all pretty good as a start, but you *can'* use dmesg. Something else
> >has to be used.
> 
> AFAIK the kernel does not store the locators anywhere - they are
> printed out by the driver but that is it.  _If_ the locators were
> stored in kernel (ala the locator names) then I would happily use
> them.  I consider parsing dmesg output an ugly hack at best.

Yes- but maybe the locator info should be retained somewhere?

> 
> > You also have to decide which is more important- spitting out
> >a config file (logical hierarchy) or printing out the device tree (hardware
> >hierarchy + pseudo device list).
> >
> 
> No, I disagree.  The routines in prtconf that gather the data about
> the devices can be shared with another tool that emits a config file,
> if you think about it the data is the same it is just the presentation
> that varies.

Hmm- yup. You're right. Too little sleep on my part. I think what I meant is
how you can reconcile a hardware tree (as gathered from incomplete
self-identifying h/w- OBP trees or PCI bussses, etc..) with a locator based
tree (which just tells you the attached driver hierarchy).

It may not be important to do so- but it would be nice also.