newbus flaw

[Doug Rabson]

On Wednesday 12 May 2004 21:30, Dag-Erling Smørgrav wrote:
> John-Mark Gurney <gurney_j at efn.org> writes:
> > You're always going to be a child of nexus, and since I assume from
> > the comment that there can only ever be one child.  Also, why do
> > you find_child w/ unit number 0, but then add a child with unit -1?
> >  Why not add it unit 0, and make it fail if that already exists?
>
> just didn't think about it.  the documentation is somewhat lacking,
> so some of the code is based on examining existing code and headers
> and guessing at what it all means.
>
> > Also, it seems to me that if dev already exists, that you shouldn't
> > reset the driver and desc.  This should be harmless, but if for
> > some reason you are called on an attached device, it could cause
> > problems.
>
> it'll all go pear-shaped if you don't.  if an ichwd device already
> exists, it is a leftover from a previous module load / unload cycle
> and the driver_t it references no longer exists.  ichwd_identify()
> should probably KASSERT that the device it finds isn't attached - I'm
> pretty sure it can't happen.

When the old module unloaded, its driver will have detached from the 
device which it created. There is no reference to an old driver_t. Its 
perfectly safe for the new driver to use the old device.